Word Order and Constituent Structure in German

CENTER FOR THE STUDY
OF LANGUAGE
AND INFORMATION
CSLI
Lecture Notes
NumberS
WORD ORDER
AND CONSTITUENT
STRUCTURE
INGERMAN
Hans Uszkoreit
CSLI was founded ea.rly in 1983 by researchers from Stanford University,
SRI International, and Xerox PARC to further research and development
of integrated theories of language, information, and computation. CSLI
headquarters and the publication offices are located at the Stanford site.
CSLIjSRI International
333 Ravenswood Avenue
Menlo Park, CA 94025
CSLljStanford
Ventura Ha.ll
Stanford, CA 94305
CSLIjXerox PARC
3333 Coyote Hill Road
Palo Alto, CA 94304
Library of Congress Catalog Card Nuruber: 87-70'215
ISBN 0-937073-09-1 (Cloth)
ISBN 0-937073-10-5 (Paper)
Copyright @1987
Center for the Study of Language and Information
Lela.nd Sta.nford Junior University
Printed in the Uniteel Sta.tes
Preface
This book is based on my Ph.D. thesis for the University of Texas at
Austin. I thank the members of my dissertation committee Lauri
Karttunen, Lee Baker, W.P. Lehmann, Stanley Peters, and Elaine Rich
for their support and patience. Special thanks go to Lauri Karttunen
who has been extremly stimulating as a teacher and colleague and very
supportive as my thesis supervisor. I have also benefitted a great deal
from interaction with Stanley Peters in classes, discussions, and joint
research.
Large parts of the research were conducted at the Center for the
Study of Language and Information at Stanford University. The mul-
tidisciplinary research environment at CSLI was an ideal setting for the
enterprise.
I would like to thank very much the people in the Artificial Intel-
ligence Center at SRI where most parts of the book were written. They
have created a very stimulating and pleasant atmosphere for both
research and writing. Let me mention here only a few: John Bear, Bar-
bara Grosz, Fernando Pereira, Jane Robinson, Stuart Shieber, and Susan
Stucky who have helped me through advice or through collaboration in
research groups.
It is impossible to mention all the other people who have influenced
the book through personal discussions at The University of Texas, at the
Stanford University Linguistics Department, at the Center for Advanced
Study in the Behavioral Sciences, at SRI, and at the Center for the Study
of Language and Information. Among them are Manfred Bierwisch,
Robin Cooper, Elisabet Engdahl, Oliver Gajek, Irene Heim, Charles
Kirkpatrick, Ivan Sag, Dieter Wunderlich, and Annie Zaenen. I also
want to thank everybody who supplied me with corrections of earlier
drafts, especially Manfred Pinkal and Peter Sells. I have used parts of
the book in courses at Stanford and in Munich. I am grateful to the
students in these classes, who gave me further comments and corrections,
If I had been able to follow all the advice I received, the outcome would
have become much better.
Thanks to the faculty and students of the Linguistics Department at
The University of Texas who have not only taught me about language
and languages but who have also helped to make my time in Austin a
wonderful and unforgettable experience. This stay would not have been
possible without the educational preparation I received from my linguis-
tics and computer science teachers in Berlin and without a two year
financial support from the German Fulbright Commission.
v
VI
My wife, Swanni Rusch, has supported me in such a variety of ways
that it would be inappropriate to list all of them here. Just to mention a
few, she has performed as an informant and critic and has assisted me in
typing and editing. I also need to acknowledge the moral support I
received from being with my son Jakob. When I started with the
research, he was not even born. When I finally got around writing it all
down, his mastery of German had gone far beyond the fragment that is
analyzed in the book.
Parts of the research for -the book were made possible by a gift from
the Systems Development Foundation and carried out in cooperation
with CSLI Stanford. Other parts were funded by NSF Grant No.
IST-8307893.
Savel Kliachko my technical editor at SRI, who edited parts of the
book, not only gave me invaluable advice regarding English style, but
was also able to detect typos in the German examples. Dikran
Karagueuzian, the CSLI editor, directed the production process with a
very supportive mix of patience and insistence. The task of producing
the book was greatly eased by editing, formatting, and printing on com-
puter facilities at SRI and CSLI.
Stanford
December, 1986
Contents
Introduction 1
1 Issues and Results 1
2 Some Theoretical Premises 3
3 Theoretical Context 5
4 The Structure of the Book 6
1 Basic Facts About German Word Order 8
1 Introduction 8
2 The Position of the Finite Verb 9
2.1 Three Clause Types 9
2.2 Verb-Second Clauses 9
2.3 Verb-Initial Clauses 11
2.4 Verb-Final Clauses 13
2.5 Summary of the Three Clause Types 13
3 The Order Within the Verb Group 15
4 The Order Within the Field of Verbal Arguments
and Free Adjuncts 18
5 The First Position in Verb-Second Clauses 24
6 Constituents Outside the Clause Core 28
2 The Framework of GPSG 31
1 Introduction 31
2 Basic Outline of the Theory 32
3 The Object Grammar 32
4 The Metagrammar 34
5 The Semantics of GPSG 38
6 Notes on the Generative Power of GPSG 39
7 GPSG and Word Order Freedom 44
Vll
Vlll
3 The Order of Main and Auxiliary Verbs 48
1 Introduction 48
2 The Position of the Main Verb 49
2.1 Basic VP Rules and Flat Clauses 49
2.2 Apparent Subject-Object Asymmetries 53
2.3 Features and Rules 60
2.4 Notes on the Semantics of VP and Clause Rules 63
3 Auxiliaries and Modals 65
3.1 Auxiliary VP and Clause Rules 65
3.2 Some Consequences and Results of the
Auxiliary Rule 68
3.3 Main Verb Analysis Versus Flat AUX Node 71
3.4 Some Notes on the Semantics of the
Auxiliary System 74
4 Verb-Second Clauses 75
4 Separable Prefixes 81
1 Introduction 81
2 The Rules 83
3 Alternative Proposals 88
4 Separable Prefixes and Topicalization 100
5 Conclusion 107
5 The Order of Verbal Complements and Adjuncts 113
1 Introduction 113
2 The Modified Framework 115
3 Discussion of the Analysis 119
3.1 Discussion of the General Approach 121
6 Evaluation and Expendability of the Grammar 128
1 Introduction 128
2 Properties of the Grammar 129
2.1 Basic Theoretical Properties of GFG 129
2.2 Generative Capacity and Implementability 130
3 Expandability 132
3.1 Additional Verb Sub categorization Frames 132
3.1.1. The Specification of the Unmarked Order 133
3.1.2. Subjectless Verbs 143
3.2 Adverbial Phrases 145
3.3 Expletive es 147
3.4 Focus Raising 151
3.5 Additional Types of Topicalization 156
IX
7 Conclusion 161
1 A Selective Summary 161
2 Directions for Further Research 162
A Rules of GFG 164
A.l Basic Rules 164
A.2 Metarules 165
A.3 Feature Cooccurrence Restrictions 165
A.4 Linear Precedence Rules 166
Bibliography 167
Author Index 175
Subject Index 177
Introd uction
1 Issues and Results
The main problem addressed in this book is the question of whether and
how the syntax of a language with partially free word order can be
analyzed in a nontransformational grammatical framework that is highly
constrained and computationally tractable, and whether such an analysis
can yield an insightful and elegant description of the relevant data. The
language chosen for analysis is German; the grammatical framework is a
modification of the immediate dominance/linear prededence (ID/LP) ver-
sion of Generalized Phrase Structure Grammar (GPSG).
There are a number of reasons that support the choice of German as
the language to be analyzed. German exhibits an interesting interaction
of configurational and nonconfigurational syntactic properties. The lan-
guage has been described and discussed extensively in the linguistic
literature. Most of its syntactic phenomena are well attested. At the
time this research was started, no work had been done on German in one
of the relevant current nontransformational frameworks. With a native
speaker of German as the author, introspection could be substituted for
more time consuming techniques of checking relevant data wherever it
seemed necessary or appropriate.
The main reason for choosing GPSG as the framework for the
proposed analysis is the recent addition of the so-called immediate
dominance/linear precedence format to the formalism. The ID/LP ver-
sion of GPSG provides the tools for encoding grammars with different
degrees of word order freedom, while still prohibiting many types of con-
stituent permutations that do not occur in natural languages. The ver-
sion of the framework that is employed for the description of German is
a highly constrained two level phrase structure grammar with desirable
computational properties.
The application of this framework to German syntax has been fru.t-
ful. The following is a summary of the main results.
1
2 Introduction
1. A phrase structure grammar for a fragment of German is
presented that encompasses the basic syntactic structures of
the language including the word-order-related phenomena and
their interactions. The fragment exhibits-among
others-the following phenomena:
a. Word order differences between main clauses and sub-
ordinate clauses.
b. The second position of the verb in assertion mam
clauses.
c. The order among main, auxiliary, and modal verbs.
d. Separable-prefix verbs.
e. The partially free order among the arguments of the
main verb.
2. A modification of the ID/LP version of GPSG is proposed.
The modified framework allows adequate encoding of the
competing principles that determine the order of verb ar-
guments and adjuncts. The suggested interface between syn-
tax and pragmatics permits description of the interaction be-
tween syntactic and discourse-determined ordering principles.
The modifications preserve the desirable formal properties of
the framework.
3. The expandability of the grammar to other syntactic
phenomena of German is discussed. This discussion focuses
on the question of whether the grammar can be expanded
without sacrificing any of its desirable formal properties and
elegant encoding strategies. The discussion arrives at positive
perspectives for phenomena such as additional
sub categorization frames for verbs, clausal subjects and
objects, subjectless verbs, adverbial phrases, and focus rais-
mg.
4. Within the discussion of the expandability of the grammar,
special attention is given to the question of which linguistic
classes of concepts need to be utilized to express ordering con-
straints. The particular problem is the case of the
unmarked-order principle. Recent work on the topic (Lenerz
(1977)) suggests that this principle should be stated in terms
of grammatical functions such as subject and object. Most
grammars in the GPSG tradition state ordering rules in terms
of syntactic categories such as NP and VP. The grammar of
the fragment follows the GPSG tradition, but extends the or-
dering rules to refer to case features. In this discussion it is
argued that ordering rules need to refer to both syntactic
categories (including case features) and thematic roles such as
agent and theme, but not to grammatical functions.
Issues and Results
3
5. The proposed solution to the problem of separable verb
prefixes should be of general linguistic interest. The sug-
gested incorporation rules are instances of non-structure-
building lexical rules, which still need to be integrated into
the framework. The analysis, which is backed by extensive
argumentation, carries over to related phenomena in other
languages, such as English particle verbs.
2 Some Theoretical Premises
The design of the most important universal frameworks for the gram-
matical description of natural languages in the recent history of linguis-
tics has been influenced by the structure of one particular
language-English.
This influence has been described and criticized as the Anglocentric
viewpoint in modern linguistics. It is most obvious in the fields of
phonology, morphology, and syntax. Striking examples of Anglocentric
theories in syntax can be found in the area of word order variation. The
relatively fixed word order in English has surely influenced the design of
different types of phrase structure grammars that serve as components in
a number of grammatical frameworks. Therefore, the analysis of lan-
guages that exhibit a greater degree of word order variation is a chal-
lenge for every such theory.
Recent work on the extremely free linear order of constituents in
some Australian languages, starting with Hale (1981), has led to inter-
esting discussions within the theories of Government and Binding (GB),
Lexical Functional Grammar (LFG), Functional Unification Grammar
(FUG), and GPSG. One of the most extreme proposals for treating the
significant differences between the strict word order of English and the
free order in languages like Warlpiri is the assumption of two radically
different classes of languages: configurational, or X, languages and
nonconfigurational, or W*, languages (Hale (1981)). This bifurcationist
view will not be debated in this book. However, it will become obvious
from the description of the framework employed and from the analysis of
German presented that the viewpoint underlying this book does not
make the distinction between configurational and nonconfigurational
languages. It is rather assumed that there is a scale of word order
freedom, the endpoints of which are not occupied by any natural lan-
guage. The word order of Warlpiri is not totally free, nor is English
word order totally fixed. A suitable formalism for the treatment of word
order freedom should be applicable to all natural languages.
In standard IDjLP grammars, the degree of word order freedom will
correspond to the number of LP rules, which, in turn, determine the
linear order of sibling constituents. Some languages will need a larger
number of such rules, others just a few. The redefinition of the LP com-
4 Introduction
ponent proposed in this book implies a different prediction. Languages
with relatively strict word order will possess many simple LP rules; free-
word-order languages, on the other hand, will have fewer LP rules that
are much more complex. These complex ordering rules are based on the
assumption that the order in free-word-order languages can be called free
only if one restricts one's attention to syntactic ordering principles.
Proponents of several non-Anglocentric linguistic theories, especially
linguists in the school of Functional Sentence Perspective, have stressed
the fact that the perceived word order freedom in Slavic languages and in
German cannot be treated as free variation. That freedom is in fact
severely constrained by the interaction of pragmatic, syntactic, and
phonological principles. Acceptance of this interaction is another premise
underlying the analysis to be presented.
A standard strategy of Tranformational Grammar has been the
separation of so-called stylistic or scrambling transformations from the
other transformation rules. Often a stylistic component of the grammar
has been assumed, that is governed by a completely different set of
regularities. The linguist working within the framework of GPSG does
not have this option. Since he works with only one level of syntactic
representations, that corresponds to the level of surface structure in a
transformational grammar, he cannot detach one kind of ordering
mechanism from others.
Another premise follows from the two-level grammar theory of
GPSG. Since LP rules operate on phrase structure rules, only sibling
constituents can be permuted. This approach is much more restricted
than, say, transformational theories or a recent proposal by Martin Kay1
for FUG. It follows from the way the LP component is defined, that all
LP rules apply in the derivation of a PS rule, i.e., the same ordering
rules apply to all sets of sibling constituents, independently of their
parent and the rest of the tree.
Throughout the presentation of the analysis, semantics is discussed
only where it is not immediately obvious what the semantic counterparts
of the syntactic rules could be. There is some limited discussion of dis-
course roles such as topic and focus. It is indicated how their contribu-
tion to the nontruthconditional part of the meaning of a sentence can be
connected with just the right ordering variants. However, little is said
about the many interesting semantic problems that arise from the inter-
action of discourse role assignment by word order and such semantic con-
cepts as quantifier scope, definiteness, specificity, illocutions, etc.
The assumed relationship between word order and semantics rests on
the following (simplified) premise. The variations that arise through
freedom in word order belong to the larger class of permutational varia-
tions. These are syntactic relationships associating sets of sentences that
1Talk at CSLI (Stanford University) in Spring 1984.
Some Theoretical Premises 5
differ from one another only in the order of constituents and in certain
nontruthconditional parts of their meanings. All permutational variants
of a sentence share with it the expressed proposition. Excluded from this
generalization are preferences in the resolution of quantifier scope am-
biguities and anaphora binding.
The last premise to be mentioned here is concerned with the criteria
that are applied to the formal properties of the grammar formalism. It
has been demonstrated by Uszkoreit and Peters (1983) that an un-
restricted GPSG has Turing machine power. The restricted version of
GPSG that is described in Gazdar and Pullum (1982a) and that is
employed in many recent applications of the framework permits gram-
mars for context-free languages only. There exists strong evidence that
at least some natural languages are not context-free (Culy (1984), Shieber
(1984)). On the other hand, it seems that the additional power needed
does not go far beyond context-freeness. Exactly how much additional
power is needed is currently an open question.
Another open question is whether a classification based on the
proper-inclusion classes of the Chomsky hierarchy can be an appropriate
measurement of the computational complexity of natural language syn-
tax. There are indications that it cannot. So far, linguists have not been
able to define or even agree on a useful characterization of necessary and
sufficient complexity. When the grammar to be presented here starts out
from the version of the framework that is constrained to context-freeness,
this move does not imply that the particular constraint is linguistically
justified. In fact, we have argued elsewhere that it is not (Uszkoreit and
Peters (1983), Shieber et al. (1983a)). Moreover, it is not known whether
Standard German is a context-free language. However, the strategy is to
start with the artificially constrained version of GPSG and then to show
that neither of the proposed rules nor the redefinition of the formalism
will require any additional generative power. By following this strategy
it can be demonstrated that, independently of any future revelations
regarding the computational complexity of German and other natural
languages, the presented analysis of German word order, along with t.he
proposed changes in the framework, does not increase the generative
power and thus keeps the formalism computationally tractable.
3 Theoretical Context
The purpose of this short section is to list selected of publications that
have influenced the analysis of German and the extensions to t.he
framework either directly or indirectly.
The research was influenced by work in all contemporary major l:in-
guistic theories. In the tradition of generative transformational gram-
mars, it was mainly early work by Bach (1962) and Bierwisch (1971) that
had an impact on all subsequent research on German syntax. Craig
6 Introduction
Thiersch's dissertation (1978) analyzes a number of problems in German
syntax in the framework of GB.
The first major GPSG for a language with relatively free word order
is that of Stucky (1981). GPSG research on word order entered a new
phase with the introduction of the ID/LP formalism (Gazdar and Pullum
(1982b )). Some preliminary results of this research are reported in
Uszkoreit (1982a; 1982b; 1983). An important and comprehensive piece
of work on a free word order language in LFG is found in Simpson
(1983). There is only published application of FUG to a language with
more word order freedom than English-Karttunen and Kay's
(1983) paper on Finnish.
The Prague group has been carrying on the theory of Functional
Sentence Perspective (Sgall, Hajicova, and Benesova (1973)). Their cur-
rent framework has its roots in Dependency Grammar. Montague gram-
mars for fragments of German varying in size can be found in von
Stechow (1978), Jacobs (1982) and Zaefferer (1984). Of special impor-
tance for the empirical part of this book were the following books and
papers: Lenerz (1977), Wunderlich (1983a; 1983b), and Heidolph et al.
(1981).
Readers whose bibliographic curiosity has not yet been sated, should
consult Etzensperger (1979) and Scaglione (1981).
4 The Structure of the Book
Chapter 1 provides a descriptive account of the syntactic phenomena
that are exhibited by the fragment. This chapter serves the sole purpose
of familiarizing the reader with well-attested data. New observations are
discussed in later chapters in connection with the proposed analysis.
Chapter 2 offers an overview of the framework of GPSG, concentrat-
ing on the ID/LP version of the formalism.
The grammar for the fragment of German is presented in Chapters
3, 4, and 5. Chapter 3 describes the analysis for the positions of main
and auxiliary verbs. It starts out with a specification of the fragment.
Subsequent sections introduce rules for verb-initial and verb-final clauses,
for auxiliary and modal verbs, and for verb-second sentences.
Chapter 4 is dedicated to the notoriously difficult problem of
separable prefixes. There it is argued that an analysis is required that
employs non-structure-building lexical rules. The solution is compared
with related proposals. Its theoretical relevance for determining the bor-
derline between syntax and morphology is discussed. It is also shown
that the analysis avoids a relaxation of the Lexical Integrity Hypothesis.
Chapter 5 deals with the linear order of the main verb's arguments.
Whereas the preceding two chapters had to account for the coexistence of
different word orders in the same language, Chapter 6 provides a gram-
mar for that part of German that exhibits the highest degree of word
The Structure of the Book 7
order freedom. The LP component of GPSG is redefined to account for
the interaction of several constraints on the ordering freedom. It is
described how the rich feature system can serve as the interface between
syntax and pragmatics.
Chapter 6 discusses the essential properties of the analysis. The dis-
cussion then turns to the question of whether these properties could be
preserved if the grammar were to be extended to wider coverage. Pos-
sible extensions to a number of additional phenomena are investigated.
Among the latter are clausal subjects and objects, subjectless sentences,
adverbial phrases, and focus raising.
The conclusion sums up the principal results and suggests directions
for further research.
Chapter 1
Basic Facts About German Word Order
1 Introduction
The purpose of this chapter is to provide the reader with the necessary
background for the discussion of the GPSG analyses in Chapter 3. The
present chapter suffers somewhat from the well-known but unavoidable
difficulties that arise when linguistic data are introduced separately from
their analysis. To reduce these difficulties to a minimum, many details
have been omitted that will be included later in Chapters 3 and 6. I
have also tried as much as possible to present the data in this chapter in
a theory-independent fashion.
A particular characteristic about German is its different word order
in main clauses and subclauses. Actually, there exist three clause types
with respect to the position of the finite verb in the clause core.
2
These
are described first. Then the order among the nonfinite verb forms and
among the other constituents of the sentence is discussed. The chapter
closes with a description of constituents that occur outside the clause
core, such as extrapositions.
2The meaning of the term clause core will be defined negatively in Section 6 by listing those
elements that do not belong to the clause core.
8
Introduction
9
2 The Position of the Finite Verb
2.1 Three Clause Types
German sentences have tense.
3
The tense-marking morphemes are verb
affixes. Every German clause has exactly one finite top-level element,
which can be a main verb, a modal, an auxiliary, or a conjunction of
such elements. Traditionally, three ordering types of German sentences
are distinguished with respect to the position of the finite element in the
clause core: verb-initial, verb-second, and verb-final clauses.
It is necessary to use the concept of clause core or some equivalent in
its characterization, for there might be additional material preceding or
following the clause core. The order of the verb-second clause type has
often been called basic order,4 since most main clauses as well as a num-
ber of subordinate clauses follow this pattern. But, before I plunge any
deeper into the bottomless maelstrom of German clause structure, I
would like to give a brief characterization of the functional distribution
of the three sentence types.
In a typical assertion main clause, many arguments of the verb, in-
cluding the subject, can precede the finite verb. In addition, adverbs or
free adverbial phrases can occupy the first position. The full range of
constituents that qualify for the first position and the pragmatic con-
straints that govern the choice of the fronted element are discussed in
more detail in Section 5; Chapter 3, Section 4; and Chapter 6, Section
3.5.
2.2 Verb-Second Clauses
Examples (la) and (lb) are main clauses with the finite verb in second
position:
3Exceptions can be found in some imperatives.
Immer auf den Verkehr achten!
always at the traffic pay-attention
Do watch the traffic always!
Other exceptions occur in dialects. The following tenseless sentence is acceptable in
Berlin dialect:
Der Olle seene Joren vakloppen, det ham wa jerne.
The old his kids thrash this have we like
That the father beats up his children, we really like that. (Ironic)
4This descriptive use of basic order must not be mistaken for the way the term is used by
typologists or within transformational grammar.
10 Basic Facts About German Word Order
(1)
a. Du kornmst morgen.
PRES-2S
You come
b. Morgen
tomorrow.
kornmst duo
PRES-2S
Tomorrow come you.
As in English, these assertion clauses can also be used to express
questions or orders:
(2) Du kornmst morgen?
(3) Du kornmst morgen!
Main-clause constituent questions (with the possible exception of echo
questions) are another class of sentences that exhibit verb-second order.
(4) Wer kornmt morgen?
PRES-3S
Who comes tomorrow?
(5) Warm kornmst du?
PRES-3S
When come you?
In constituent questions-again with the possible exception of echo
questions-the first position is occupied by a phrase containing the inter-
rogative element.
Although the finite verbs of subordinate clauses are usually clause-
final, there are cases in which the sentential complements of certain verbs
are in verb-second order.
(6) Ich weiB, du kornmst morgen.
I know you come tomorrow.
The phenomenon extends to subordinate questions.
(7) Ich frage mich, warm kornmst duo
I ask myself when come you.
These verb-second subordinate clauses carry a quotation character
analogous to their English counterparts in (8a) and (9a) (in contrast to
(8b) and (9b)).
The Position of the Finite Verb
11
(8)
a. I know, you're coming tomorrow.
b. I know that you're coming tomorrow.
(9)
a. I wonder when does he come.
b. I wonder when you are coming.
The quotation character is not present when the verb-second sub-
ordinate clause uses the first subjunctive to indicate indirect (reported)
speech.
(10) Peter sagte, er komme morgen.
SBJl
Peter said he come tomorrow
Peter said that he would come tomorrow.
2.3 Verb-Initial Clauses
Alternative questions in German, just as in, start with a finite verb form
when they are main clauses.
(11) Kommst du morgen?
come you tomorrow
Will you come tomorrow?
Syntactic imperatives-as distinct from other syntactic clause types,
which are used to convey orders or requests-are also verb-initial.
(12) Kommmorgen.
IMP
Come tomorrow.
(13) KommenSie morgen.
IMP(formal)
come you tomorrow
Come tomorrow!
Antecedents of conditionals can be marked either by particles Like
wenn or falls (if or in case) or by the initial position of the finite verb.
(14) Wenn Peter morgen kommt, (dann) komme ich schon heute.
if Peter tomorrow comes (then) come I already today
If Peter comes tomorrow then I shall already come today.
(15) Kommt Peter morgen, (dann) komme ich schon heute.
comes Peter tomorrow (then) come I already today
If Peter comes tomorrow then I shall already come today.
Verb-initial clauses can also be found in interjections.
(16) Kommst Du doch wieder zu spat!
come you again too late
Oh., you are too late again!
(17) 1st das ein Sauwetter!
is this a sow-weather
This is really terrible weather!
Finally, there is at least one more class of verb-initial sentences:
subjunctive sub clauses introduced by ale, as in (18b) and (19b).5
(18)
a. Er benimmt sich, als ob er selten ptinktlichkomme.
SBJ1
he behaves self as if he seldom on time come
He behaves as if he seldom came on time.
b. Er benimmt sich, als komme er selten ptinktlich.
SBJ1
he behaves self as come he seldom on time
He behaves as if he seldom came on time.
(19)
a. Sie behandelten ibn, als wenn er nie ptinktlich
they treated him as if he never on time
kame.
SBJ2
came
They treated him as if he never came on time.
b. Sie behandelten ibn, als kame er nie ptinktlich.
SBJ2
they treated him as came he never on time
They treated him as if he never came on time.
Sentences (18b) and (19b) are considered examples of verb-initial clauses,
even though the verb follows a subordinating conjunction, because in
5The examples are variations of Heidolph et al. (1981), p. 718, Abb. 21.
The Position of the Finite Verb 13
most analyses the subordinate clause (~nexpands to a subordinate con-
junction (or some other complementizer) plus a clause (S).
2.4 Verb-Final Clauses
Subordinate clauses introduced by a conjunction, a complementizer, or a
phrase containing an interrogative or relative item are verb-final. Excep-
tions are illustrated in (18b) and (1gb) in Section 2.3.
(20) Peter kommt, nachdem Paul nach hause geht.
Peter comes after Paul to home goes
Peter comes after Paul goes home.
(21) Paul weiB, daB Peter nach hause kommt.
Paul knows that Peter to home comes
Peter knows that Peter is coming home.
(22) Paul weiB, wer in die Schule kommt.
Paul knows who to the school comes
Peter knows who is coming to school.
(23) PaulXennt den Mann, der zu spat kommt.
Paul knows the man who too late comes
Paul knows the man who is late.
2.5 Summary of the Three Clause Types
With the exception of the position of the finite verb, I have not yet said
anything about German word order as such. Fortunately, there is a very
natural way of decomposing the problem further. To this end, let me
introduce the following four abbreviations: F will stand for the first ele-
ment in verb-second clauses; A will be an abbreviation for syntactic ar-
guments of verbs, including the subject, and for any free adjuncts; V
F1N
will be the finite verb; V
1NF
symbolize any nonfinite verb form.
Strictly for expository purposes, without any claims being made with
regard to the constituent structure, the clause core strings for the three
clause types can be represented as follows.
(24)
a. F V
FIN
A* V;NF
b. V
FIN
A* V;NF
c. A* V;NF V
FIN
The asterisk stands for the Kleene star as it is used in regular expres-
sions. Thus, the verb-second clause consists of some undetermined first
constituent, followed by a finite verb form, in turn followed by zero or
more arguments or adjuncts, finally followed by zero or more nonfinite
verb forms.
Only F and V
FIN
are obligatory. F might be the subject of an in-
transitive verb, as in (25), or the sole argument of a subjectless verb'' as
in (26).
(25) Peter kommt.
Peter comes.
(26) Ihn friert.
ACC 3S
him freezes
He is cold.
An example with one A and V
1NF
each is (27).
(27) Das Paket hat die Post gebracht.
ACC PRES-3S NOM PSP
the parcel has the mail brought
The parcel has come in the mail.
A verb-initial clause with a finite main verb and without any overt
arguments is the imperative in (28).
(28) Komm!
Come!
Verb-final sub clauses, as notated in (24c), are still incomplete because
they have to be preceded by a relative phrase to form a relative clause or
by a subordinating conjunction to form other types of sub clauses. A*
can be empty only if the subject (or the obligatory argument of a sub-
jectless clause) has been relativized as in (29).
(29) Der Mann, der
NOM NOM
schlift, schnarcht.
PRES-3S
The man who sleeps snores.
The clause type schemata in (24) suggest the following useful sub-
division of the word order phenomena not yet discussed: the order within
6Subjectless verbs will be discussed in more detail in Chapter 6, Section 3.1.2.
The Position of the Finite Verb 15
A* (i.e., the ordering of subject, objects, other complements, and
adjuncts), the order of the nonfinite verb forms, and the nature of the
first element in verb-second sentences. I shall discuss each of these sub-
problems briefly. A section on those constituents that can be found out-
side the clause core will conclude this chapter.
3 The Order Within the Verb Group
In English, auxiliaries and modals occur in fixed order. In (30) no per-
mutations are allowed within the verb group.
(30) Peter could have been seen.
The German translation of (30), put in the verb-final subclause or-
der, shows a mirror image of the English verb sequence:
(31) ...weil Peter gesehen worden sein konnte.
because Peter seen been have could
... because Peter could have been seen.
In the corresponding main clause, the finite verb form occupies
second position. The sequence within the nonfinite verbs remains the
mirror image of the corresponding sequence in English:
(32) Peter konnte gesehen worden sein.
Again, in both (30) and (32), no permutations are possible within the
verb group. However, there are a number of differences between German
and English auxiliary strings; at this point I would like to mention just a
few of them.
There is no progressive aspect marking like the one achieved in
English by combining the auxiliary be with the suffix -ing.
(33) *Peter ist suchend.
Peter is searching.
The present-participle verb forms are restricted to adjectival use, as
in (34a) and (34b), to nominal modifiers, as in (34c), and to adjuncts, as
in (34d).
(34)
a. Peter ist fordernd.
Peter is demanding.
b. Paul ist noch fordernder.
Paul is even (more) demanding.
c. Der das Buch suchende Mann ist Peter.
the the book searching man is Peter
The man searching for the book is Peter.
d. Peter lief umher, das verlorene Buch suchend.
Peter ran around the lost book searching
Peter ran around, searching for the lost book.
There can be more than one modal in the verb group, as in (35).
(35) Peter will nach Hause gehen durfen.
Peter wants to home go may
Peter wants to be allowed to go home.
Will is a modal and has semantic scope over the other verbs. Example
(36) is also a syntactically well-formed German sentence.
(36) Peter darf nach Hause gehen wollen.
Peter may to home go want
Peter is allowed to want to go home.
This time the modal dar f has widest scope.
Just as in certain Romance languages, there are two auxiliaries in
German that mark the perfective aspect of verbs: haben and sein. The
choice depends on the individual main verb.
(37)
a. Peter hat geschlafen.
Peter has slept.
b. *Peter ist geschlafen.
Peter has slept.
(38)
a. *Peter hat gelaufen.
Peter has run.
b. Peter ist gelaufen.
Peter has run.
Obviously, German modals need to have nonfinite as well as finite
forms to allow for double modals. This enables them to occur as well
between the main verb and the future auxiliary werden (39), or between
the main verb and the perfective auxiliaries haben or sein (40).
(39) Peter wird kommen durfen.
Peter will come may
Peter will be allowed to come.
The Order Within the Verb Group 17
(40) Peter hat kommen durfen.
Peter has come may
Peter has been allowed to come.
Many German verbs have, i.e., prefixes that seem to form a word
together with the verb in some cases and that are separated from its
stem in others. One class of these prefixes resembles the English verb
particles like up in (41).
(41) Peter picks the mail up.
The German translation in (42) shows the prefix separated from the
verb, just like the English particle in (41).
(42) Peter holt die Post abo
Peter picks the mail up.
It is only when the main verb assumes the first or second posrtron that
the particle is separated from the stem. In all other cases, including
deverbal members of other categories, the prefix appears as in (43).
(43)
a. Peter kann die Post abholen.
Peter can the mail up-pick
Peter can pick the mail up.
b. Die Abholung ist urn 3 Uhr.
The pickup is at 3:00.
The position of the separated prefix is identical to the position of t.he
nonfinite verb group, i.e., at the end of the clause core. The (almost
complete) complementary distribution of separated prefix and nonfinite
verb string derives from the fact that the prefix can be separated only if
the main verb itself is finite. Separable prefixes will be discussed in
greater detail in Chapter 4.
The following schema summarizes the most relevant facts about the
order of the nonfinite verb string if it is neither empty nor formed by a
separated prefix.
(44) MY(PA) MA* (AA) MA* (FA)
MY- main verb
PA - passive auxiliary
MA- modal auxiliary
AA - aspectual aux.
FA - future tense aux.
(geben)
(werden)
[konnen}
(haben)
(werden)
Again, the asterisk denotes the Kleene star. The semantics of modals
effectively restricts the actual occurrence of multiple-modal sequences to
a very small number. If one of the verb forms in this string is finite, it is
always the rightmost one.
4 The Order Within the Field of Verbal Arguments and
Free Adjuncts
This section will examine an area of interest with respect to German
word order in which a high degree of ordering variation is exhibited.
When people talk about free word order in German, they most likely
mean the permutation that is possible within the string of complements
and adjuncts of the verb. I shall defer until later any discussion of the
first position of verb-second sentences, focussing instead at this point on
another issue-the contiguous sequence of nonverbal elements in the sen-
tence.
If this sequence is formed by the arguments of a ditransitive
verb-subject (SUBJ), direct object (DOBJ), and indirect object
(IOBJ)-then all six permutations of these elements may give rise to ac-
ceptable sentences (45).
(45)
a. Dann wird der Doktor dem Patienten die Pille geben.
then will the doctor the patient the pill given
b. Dann wird der Doktor die Pille dem Patienten geben.
c. Dann wird die Pille der Doktor dem Patienten geben.
d. Dann wird dem Patienten der Doktor die Pille geben.
e. Dann wird dem Patienten die Pille der Doktor geben.
f. Dann wird die Pille dem Patienten der Doktor geben.
Then the doctor will give the patient the pill.
All permutations have the same truth-conditional meaning.
However, not every permutation of SUBJ, DOBJ, and IOBJ might be ac-
cepted as grammatical. Consider (46a), which is a permutation of (46b):
(46)
a. *Dann hatte einen groBen Spielzeuglaster ihm es
Then had a big toy truck him it
gegeben.
given.
Then it had given him a big toy truck.
The Order Within the Field of Verbal Arguments
19
b. Dann hatte es ihm einen groBen Spielzeuglaster
Then had it him a big toy truck
gegeben.
given.
All other grammatical permutations of the nominal elements of (4Gb)
are judged with respect to their acceptability somewhere between (4Gb)
and (46a). An important property of the ordering variation is the con-
textual dependence of permutation variants. A variant that appears ap-
propriate in one context might sound awkward in another. Word order
variability in German shares this property with many, if not all, free
word-order phenomena across languages. The investigation of contextual
dependence has led to interesting theoretical results for Hungarian,
Japanese, and many Slavic languages, nor is the phenomenon lacking in
English. Sentences (47a) and (47b), which are related by permutation
and share the same propositional content, can differ in appropriateness,
depending upon their respective specific content.
(47)
a. I'll talk to you over lunch about the proposal.
b. I'll talk to you about the proposal over lunch.
Sentence (47a) would be more appropriate as an answer to (48a,),
while (47b) could be a response to (48b).
(48)
a. Why are you asking me to have lunch with you?
b. When can I have your comments?
The two phrases that are permuted in (47), over lunch and about the
proposal, are both adjuncts. In (48a) lunch has been mentioned. In the
reply (47a) the talk about the proposal constitutes new information. In
the sequence (48b) and (47b), the opposite is the case. Here the question
is obviously about the proposal and the time of discussion is added as
new information in (47b). There seems to be a tendency for old infor-
mation to precede new information. This is also a regularity observed in
German, where it extends to all arguments of verbs and to adverbial
phrases as well. If one hears (49) below, one can deduce with high prob-
ability prior knowledge regarding the existence of a doctor and a pill, but
one could not deduce such prior knowledge of the fact that there would
be a man to whom the pill would be given.
(49) Dann wird der Doktor die Pille einem Patienten geben.
then will the doctor the pill a patient give
Then the doctor will give the pill to a patient.
Here too the last element in the permutable string expresses part of
the new information. But now let us consider the following silly sequence
of sentences about a pill needed for different reasons by a patient and by
a detective.
(50)
a. Der Doktor hat nur noch eine Pille zu vergeben.
The doctor has only one pill left to give out.
b. Um diese Pille bitten ein Patient und ein Detektiv.
A patient and a detective are both asking for this pill.
c. Dann wird der Doktor die Pille dem Patienten geben.
then will the doctor the pill the patient give
Then the doctor will give the pill to the patient.
In this context dem Patienten in (50c) which is identical to our pre-
vious example (45), does not refer to an individual that is new in the
context. This is made even clearer by the choice of the definite article.
Simplistic views about new information following old information will
not suffice. There are two solutions to this apparent problem; either
another regularity must be stated (50c) or the concept of new infor-
mation must be refined. Both strategies have been pursued. The ad-
ditional regularity could be stated as: relevant information follows less
relevant information. The refined concept of new information would af-
firm that elements of new information follow elements of old infor-
mation. In (50c), the new information would be dem Patienten geben
(give to the patient), although both the act of giving and the patient
have been mentioned previously.
The inherent difficulty of ascertaining exactly what should be the
correct characterization of the dichotomy underlying the tendency of
phrases that denote new, relevant information to follow other phrases has
led to a multitude of informal definitions. Some of these attempt to
create such complex concepts as psychological subject and psychological
predicate, while others break the problem down into two or more
dichotomies, e.g., mentioned/new and known/unknown. Most ap-
proaches that seek to explain ordering regularities among complements
and adjuncts do not ascribe responsibility to pragmatics alone for the
detected preferences and restrictions.
In addition to pragmatically based regularities, syntactic and phono-
logical ordering principles have been suggested. The most interesting
syntactic principle is the so-called unmarked order. It has been claimed
that there is a preferred or unmarked order that has to be adhered to
unless sufficiently cogent pragmatic or stylistic factors compel violation.
The assumption is not new; many generations of students of German as a
foreign language have benefited from the principle of unmarked order as
a helpful guideline.
Unfortunately, most arguments in defense of the principle have been
based largely on intuition. Lenerz (1977) breaks with this tradition by
designing a small number of easy tests. As for the sequence of SUBJ,
DOBJ, IOBJ, Lenerz finds evidence for the unmarked order: SUBJ,
IOBJ, DOBJ. He extends this principle to include different types of ad-
verbial phrases as well.
The principle of unmarked order explains why the following sentence
appears well-formed even when it appears in a context where das Lied, a
certain song, has been mentioned before-whereas ein Kind, some un-
specified child, is part of the new information.
(51) Gestern hatte ein Kind das Lied gesungen.
yesterday had a child the song sung
Yesterday a child had sung the song.
Here the subject precedes the direct object.
Lenerz's tests also correctly predict that violation of the unmarked
order, together with violation of the pragmatic principle, results in far
less acceptable sentences. In Sentence (52) the direct object precedes the
subject, even though it refers to a part of the new information; the sub-
ject itself refers to a known individual.
(52) ?Gestern hatte ein Lied dieses Kind gesungen.
yesterday had a song this child sung
Yesterday this child had sung a song.
The assumption of an unmarked order and Lenerz's findings ob-
viously raise more questions than can be answered in this overview.
Some of these issues, however, will be discussed later in Chapter 6, SEC-
tion 3.1.1.
More syntactic principles governing the order of complements and
adjuncts have been suggested. It has been observed that definite noun
phrases usually come before indefinite ones. Personal pronouns show tile
tendency to precede nonpronominal noun phrases. In Chapter 5, Section
3, I shall come back to the question whether these syntactical principles
can be explained in terms of the aforementioned pragmatic regularities.
The last group of ordering principles I would like to mention is based
on the phonological structure of the sentence. There is a well-known
phenomenon in English called Heavy-NP-Shift. Ross (1967) labels noun
phrases as heavy if they exhibit a high degree of internal complexity, i.e.,
if they contain conjunctions or complex modifiers. The movement trans-
formation of Heavy-NP-Shift was designed to account for the fact that
some heavy noun phrases, such as object NPs, can occur at the end of
the sentence even if the resulting sequence departs from the standard or-
der. The heavy object noun phrase in (53b) follows the prepositional ob-
ject.
(53)
a. Peter took the airgram to the office.
b. Peter took to the office all the large parcels he had
packed the night before.
From this example and the first characterization of the phenomenon
it does not follow that heaviness is a phonological concept. It could also
refer to the complexity in syntactic structure or to the amount of seman-
tic information. However, to the native speaker the deviant position of
the object in (54a) seems to be less acceptable than its counterpart in
(54b) although the two object noun phrases differ only with respect to
the lengths of their words.
(54)
a. Peter saw at the party his rich friends.
b. Peter saw at the party nurnerous
industrial entrepreneurs.
In German there is also a tendency for heavier phrases to follow
lighter ones. Following Behaghel (1932), this principle has been called
das Gesetz der wachsenden Glieder (literally: the law of the growing
constituents). This law which simply states that a shorter constituent
precedes a longer one, if possible, accounts for the contrast between (55a)
and (55b).
(55)
a. ?Ich erklarte die Relativitatstheorie den Kindem.
I explained the theory of relativity the children
I explained the theory of relativity to the children.
b. Ich erklarte die Relativitatstheorie den Kindem
I explained the theory of relativity the children
in meiner Klasse, die die notwendigen Vorkenntnisse
in my class who the necessary knowledge
haben.
have
I explained the theory of relativity to those children
in my class who have the necessary knowledge.
A phonologically conditioned ordering principle often omitted from
descriptions of German word order is based on the rhythm of the sen-
tence. Behaghel (1932) formulates it as a striving for tonal dynamics.
This principle might not be as strong as those previously mentioned, but
it can affect the acceptability of borderline sentences. Many native
speakers will prefer (56a) over (56b) when they are asked to judge their
respective acceptability, although the perceived difference is rather sub-
tle. The primary accent in the examples is indicated by', secondary ac-
cent by'.
(56)
a. Ich zeigte das Schauspiel Marlene.
I showed the play Marleene.
b. Ich zeigte die Parod1e Kurt.
I showed the parody Kurt.
c. Ich zeigte Kurt die Parodie.
I showed Kurt the parOdy.
The metric structure of German (as well as English) sentences is per-
ceived as superior if stress clashes as in (56b) are avoided. An obvious
alternative to (56b) is (56c).
If we assume an invariant context, the primary stress of the sentence
will remain on the focus, even if the focus is not the last element. This is
typical for the type of interaction between prosody and word order that
has not yet been discussed. If the pragmatically suggested order is
violated-usually because other principles are being observed-then the
prosodic structure of the sentence will deviate from the standard pattern.
The regularity behind this change can be seen as the manifestation of
two ways to mark the focus or comment of the sentence. It is marked
either by having it follow all nonfocus elements of the clause core (except
for the nonfinite verbs) or by assigning it a primary accent even if it is
not in accent position. This is also a well-known phenomenon in English
prosodies. In this connection, the reader is asked to recall sentence (47b),
which is repeated below in (57).
(57) I'll talk to you about the proposal over lunch.
It was asserted earlier that this sentence would be most appropriate
in a context in which it is shared knowledge that the proposal is likely
to be talked about but that the time still needs to be agreed upon. But
this claim can easily be refuted if deviations from standard (or
unmarked) intonation are allowed. If the main accent is on proposal, the
sentence would be appropriate in a context that deals with a lunch date
and in which the subject of conversation has not yet been established.
In English the range of cases in which the speaker has a choice be-
tween a marked intonation or a simple phrasal permutation is of course
much narrower than in German. Other syntactic constructions that ful-
fill functions similar to focus intonation or focus permutation are Left
Dislocation, Cleft Formation, and Left Dislocation.
Let us now summarize the most relevant regularities that constrain
the constituent order among complements of the verb and adjuncts.
There are two alternative ways to describe this partially free word order.
One is based on the premise that the order is actually fixed-e.g., by
virtue of some interaction of the pragmatic principle (focus after
nonfocus) with the unmarked order; provision could then be made for
any exceptions. The other approach is to start out with the notion of a
completely free order and then, as necessary, to impose a number of
restrictions. As one might expect, this choice has led to serious theoreti-
cal arguments. For explanatory reasons I prefer the latter alternative
here, but I shall return briefly to the theoretical question in Chapter 6,
Section 3.1.1.
The sequential order of the constituents considered here can then be
regarded as basically free. However, the permutations that do occur con-
form to certain principles (not all of which are always observed). The
most relevant principles are these:
Focus follows nonfocus
The unmarked order is SUBJ, IOBJ, DOBJ.
Personal pronouns precede other NPs.
Definite NPs precede non definite NPs.
Light constituents precede heavy constituents.
If a focussed constituent precedes a nonfocussed it will carry
the focus accent.
The interaction of the principles will receive more attention in Chap-
ter 5.
5 The First Position in Verb-Second Clauses
It was Erich Drach (1963) who first formulated the general rule stating
that every major clausal constituent (Satzglied), with the exception of
the finite verb, can occupy the first position in verb-second clauses. Von
Stechow (1978) therefore proposed that this generalization be called
Drach's Law. It represented the first attempt to break with the tradi-
tional view that German had a basic order in which the subject preceded
the finite verb and that all clause patterns deviating from this order were
inversions of one kind or another. Exactly which strings would qualify
as Satzglieder, however, is still an open question. I shall investigate this
problem in some detail in Chapter 3, Section 4 and Chapter 6, Section
3.5. In this section I simply list some of the generally accepted
regularities that may be found in standard descriptive grammars of Ger-
man. These will serve as sufficient background for discussion of the
analysis presented in Chapter 3.
The first position can be occupied either by the subject or by any
object (including prepositional objects).
(58)
a. Peter schickte den Kindem das Paket.
Peter sent the children the parcel
Peter sent the parcel to the children.
b. Den Kindern schickte Peter das Paket.
the children sent Peter the parcel
or To the children Peter sent the parcel.
c. Das Paket schickte Peter den Kindem.
the parcel sent Peter the children
or The parcel Peter sent to the children.
d. Um schnelle Beforderung hatte er gebeten.
about fast transportation had he asked
He had requested fast delivery.
The first position can also be occupied by subcategorized-for free ad-
verbial phrases, adverbs, and predicatives.
(59)
a. Auf die Waage hatte er das Paket gelegt.
on the scale had he the parcel put
He had put the parcel onto the scale.
b. Am Morgen hatte er es gepackt.
in-the morning had he it packed
In the morning he had packed the parcel.
c. Auf der Post hatte er gewartet.
at the post office had he waited
At the post office he had waited.
d. Weil er warten muBte, war er nicht plinktlich
because he wait must was he not on time
gekommen.
come
Because he had to wait he did not come on time.
e. Schnell war er gefahren.
fast was he driven
He had driven fast.
f. Aergerlich war er.
angry was he
He was angry.
It could even be a nonfinite verb that precedes the finite verb, as in (60).
(60) Abgeschickt hatte er das Paket.
off-sent had he the parcel
He had sent the parcel off.
There are many phrases that cannot be fronted. They could be ex-
cluded on independent grounds from the class of major phrasal
categories. The fronted determiner in (61) does not belong to a major
category.
(61) *Das hatte er
The had he
Paket abgeschickt.
parcel off-sent.
The fronted noun phrase in (62) is a major phrasal category but it is
a major phrasal constituent of a lower (or embedded) phrase.
(62) *Peter wuBte Paul daB ein Paket abschickt.
One restriction, therefore, is that the fronted constituent has to be a
major phrasal constituent of the matrix clause. Of course, this is still a
very informal and fuzzy characterization.
In addition to the syntactic restrictions, there are also pragmatic
constraints on the set of constituents that can be fronted. For the time
being I prefer to simplify matters considerably by postulating that there
are only two conditions under which constituents qualify. The thing
denoted by the phrase might constitute the topic of the sentence. The
question in (63a) makes Peter the topic of (63b)_7 The demonstrative
pronoun dem, as it is used in (63b), corresponds to the English personal
pronoun him.
(63)
a. Und was war mit Peter?
and what was with Peter
And what about Peter?
b. Den hat die Polizei geschnappt.
him has the police caught
He was caught by the police.
7For a characterization of the notion topic refer to Chapter 3, Section 4.
The fronting of a constituent together with an emphatic accent on it
could also mark the fact that its denotation (or a part thereof) is contras-
tively sed upon. In (64b) Peter is once again the topic of the sentence.
The tax authority is contrasted with the police and thus the phrase
denoting it is fronted.
(64)
a. Und was war mit Peter?
and what was with Peter
And what about Peter?
b. Das Ff.nanzamt, hat ihn geschnappt (und nicht
the finance authority has him caught (and not
die Polizei).
the police)
It was the IRS that caught him (and not the police).
or
It was the IRS he was caught by (and not the police.)
The emphatic focus on the fronted denotation of the constituent does
not necessarily have to be contrastive. The emphasis on the fronted con-
stituent can be the expression of different kinds of emotional involve-
ment. Consider (65), which is another possible reply to (64a).
(65) Einem lausigen Schnuffler ist er in die Fange geraten.
a lousy snooper is he in the clutches got
A lousy snooper got him in his clutches.
Before I go on to the first position in constituent questions, let me
briefly summarize the restrictions on filling this slot in assertion clauses.
If a major phrasal constituent of an assertion main clause denotes either
the topic of the sentence or if its denotation receives emphatic focus, the
constituent can occupy the position before the finite verb form. This
fronting of constituents resembles English topicalization in a number of
ways. These similarities will be discussed in Chapter 3, Section 4.
In German, just as in English constituent questions, the constituent
containing the wh-item is fronted.
(66)
a. Welchen Kindem hat Peter ein Paket geschickt?
which children has Peter a parcel sent
To which children has Peter sent a parcel?
b. Was hat Peter den Kindem geschickt?
what has Peter the children sent
VVhat has Peter sent the children?
c. Wer hat den Kindem ein Paket geschickt?
who has the children a parcel sent
Who has sent the children a parcel?
There is no preposition stranding as in English.
(67)
a. *Was hatte er gebeten urn?
What had he asked for?
b. Um was hatte er gebeten?
For what had he asked?
The problem of deciding whether topicalization and constituent ques-
tion formation in English syntax can be regarded as two instances of a
more general syntactic process, i.e., Wh-Movement, is equally applicable
for German. Although I do not claim to have a solution, I refer the in-
terested reader to Chapter 3, Section 4 and Chapter 6, Section 3.5, for
more data on both phenomena.
6 Constituents Outside the Clause Core
The foregoing description of the basic facts of German word order has
been restricted to the clause core. This limitation has already proven
helpful in defining the three basic clause types. There are some con-
stituents that can precede the clause core. Among those are utterance
modifiers, as in (68a) and (68b), conjunctions (68c), focus particles (68d),
and noun phrases fronted by left-dislocation (68e).
(68)
a. Uebrigens, Peter kommt morgen.
By the way, Peter comes tomorrow.
b. Was Peter betrifft, der kommt morgen.
what Peter concerns, he comes tomorrow
As to Peter, he is coming tomorrow.
c. Und Peter kommt morgen.
And Peter comes tomorrow.
d. Nur Peter kommt morgen.
Only Peter comes tomorrow.
e. Peter, der kommt morgen.
Peter, he comes tomorrow.
The clause core can be followed by extrapositions. As in English, the
range of constituent types that can be extraposed varies among speakers.
Some examples of common extrapositions are given below. In (69a) a
relative clause is extraposed, in (69b) a that-clause, in (69c) an infinitival
The Order Within the Field of Verbal Arguments "29
verb phrase complement, in (69d) a prepositional phrase, and in (6ge) a
heavy noun phrase.
(69)
a. Morgen werden die Kinder kommen, die uns
tomorrow will the children come who us
noch nicht kennen.
yet not know
Tomorrow the children will come who do not yet know us.
b. Peter hat gewuBt, daB die Kinder kommen werden.
Peter has known that the children come will
Peter knew that the children would come.
c. Er hatte die Kinder gebeten, zu uns zu kommen.
he had the children asked to us to come
He had asked the children to come to us.
d. Morgen werden auch Kinder kommen mit
tomorrow will also children come with
erheblichen Schulproblemen.
significant school problems
Children with significant school problems will be
coming tomorrow as well.
e. In seinem Buch hat Peter vor allem beschrieben
in his book has Peter especially described
Kinder, die in der Schule gescheitert waren.
children who in the school failed were
In his book Peter described especially children who had
failed in school
In addition to extraposed elements, the German clause core can also
be followed by elliptical phrases that could be intuitively described as ex-
pressing afterthoughts (70).
(70) Peter hatte solche Kinder oft selbst getroffen,
Peter had such children often self met
wahre Opfer ihrer Umgebung.
true victims of-their environment
Peter had often himself met such children, true victims of
their environment.
These postscripts cannot be subsumed under extrapositions.
For the sake of completeness, one last type of phrase should be men-
tioned that is not part of the clause core, even though it is sometimes
enclosed therein. As in English, parentheticals can interrupt a German
sentence at many points. The same phrases can often also precede or
follow the sentence.
(71)
a. Peter, glaube ich, hatte eine Vorliebe fur lange
Satze.
Peter, I believe, had a preference for long sentences.
b. Peter hatte, glaube ich, eine Vorliebe fur lange
Satze.
c. Peter hatte eine Vorliebe fur lange Satze, glaube
ich.
Nothing else need at this time be said about. Although clearly
belonging to the phenomena of configurational variation, they do not in-
teract in any interesting way with the word order regularities so charac-
teristic of German that I discuss in this book.
Chapter 2
The Framework of GPSG
1 Introduction
Nontransformational grammatical frameworks have been on the rise for
at least the past five years. The versions of one of the most widespread
nontransformational theories are subsumed today under the name
Generalized Phrase Structure Grammar (GPSG). The framework has
grown out of pioneering work by Gerald Gazdar in the years 1978 and
1979. Today there exists a considerable number of publications that use
or discuss GPSG. (For references check Gazdar and Pullum (1982c)).
This chapter will not reiterate my reasons for choosing GPSG dis-
cussed in the Introduction. I will not be concerned with the history of
the theory nor will I try to discuss all versions of the framework that are
currently under consideration. The chapter will concentrate instead on
the general features of the framework and on the description of some
tools that are central to the analysis of German word order presented in
Chapters 4 and 5. For more extensive descriptions and definitions of the
components of GPSG, the reader is referred to Gazdar (1982a), Gazdar
and Pullum (1982a), Klein and Sag (1982), and Thompson (1982).
The chapter starts with a general outline of the framework. Then
the phrase structure component of GPSG is characterized. The com-
ponents of the metagrammar are discussed in Section 4. Another section
presents a superficial survey of the framework's semantics. Section 6
describes some recent results with respect to the power of the formalism.
The last section contains a discussion of some metatheoretical problems
that are relevant to research in the area of word order freedom.
31
32 The Framework of GPSG
2 Basic Outline of the Theory
In an unpublished paper, Gerald Gazdar (1979) challenged the widely
held claim that context-free phrase structure (CF-PS) grammars are not
powerful enough to serve as appropriate tools for the description of
natural language.
The attacked claim was based on two assumptions, either of which
alone could be used to justify the need for more powerful formalisms,
such as Generative Transformational Grammar (TG) of Chomsky (1957;
1965) or of subsequent versions. One assumption held that the set of
natural languages is not properly included in the set of CF languages.
This assumption was supported by several examples of constructions that
supposedly could not be generated by CF-grammars. Gazdar and Pul-
lum (1982d) and others have questioned the validity of the examples
given in support of this hypothesis.
The second assumption was that, even if CF-PS grammars were ob-
servationally adequate (i.e., if they were able to generate the string sets
of any natural languages) they would nevertheless still not be able to ex-
press the linguistic generalizations necessary for characterizing linguistic
competence in an illuminating manner.
The proponents of GPSG have opposed both assumptions. I will
postpone the discussion of their theoretical argumentation and will
procede instead directly to their practical arguments. The first assump-
tion can be refuted by demonstrating that natural languages-including
the constructions used as examples countering the context-freeness
claim-can in fact be generated by CF-PS grammars with finite sets of
rules that have finite length.
But such grammars might still be subject to the criticism that fol-
lows from the second assumption, for they are really not the appropriate
tool for expressing all the syntactic relations that can be embodied by
transformations. However, a grammar of this kind is only one major
component of a particular GPSG, i.e., the object grammar, which can be
utilized for generation or recognition in a straightforward manner.
The second major component of a GPSG is a metagrammar. An in-
ductive definition of the object grammar, it can be viewed as a grammar
that generates the object grammar. In a TG, syntactic relations among
sentences are reflected in their derivational history. In a GPSG, these
relations are usually reflected in the derivation of the rules that generate
those sentences.
3 The Object Grammar
The object grammar combines CF-PS syntax and model-theoretic
semantics. Its rules are ordered triples (n; r; t), where n is an integer,
the rule number, r is a CF-PS rule, and t is the tranelaiion of the rule.
The Object Grammar :33
The term rule number for n is slightly misleading, since the same integer
can occur in several rules. Element n is better described as the
subcategorization marker, for every preterminal category in the right-
hand side of r can expand only to those words in the lexicon that are
marked n.
8
Rule r is a CF-PS rule. Its nonterminal symbols are complex
categories, i.e., bundles of features. The features that mark the member-
ship of the symbol in one of the traditional syntactic categories conform
to the notation of X-syntax (Jackendoff, (1977)). The exact nature of
the feature system has been a much-disputed topic. I assume here that a
category is a set of features, although it has been suggested that more
structure should be assigned to categories (e.g., Gazdar and Pullum
(1982a)).
Translation t is a schema that expresses the denotation of the left-
hand-side constituent as a function of the denotation of its right-hand-
side constituents. It is similar to translation rules espoused by Montague
(1974a; 1974b). The target language is a version of intensional logic that
can be interpreted by using the standard apparatus of Montague-style
model-theoretic semantics.
The rules of the object grammar are interpreted as tree-admissibility
conditions in which syntax and semantics apply in tandem. Example
(72) shows a (simplified) object grammar rule for English:
This rule admits a (sub )tree of the following form:
(73) v
2
/~2
V N
The root node of category VP directly dominates a verb node and a
noun phrase node (in that order). The verb node must immediately
dominate a verb whose sub categorization marker is 23 (in this case a
transitive verb). The semantic translation of the verb phrase is the value
the function denoted by the verb acquires when it is applied to the inten-
sion of the noun phrase denotation.
8The special status of n, as opposed to the features of the left-hand-side category, will
become apparent when I describe the functioning of metarules. However, there exist
proposals to replace the rule number with a feature (Carl Pollard and Mark Gawron,
personal communication).
34
The Framework of GPSG
4 The Metagrammar
The metagrammar consists of four different kinds of rules that are used
by three major components to generate the object grammar in a stepwise
fashion.
g
The chart (74) illustrates the basic structure of a GPSG
metagrammar:
(74)
Metarule application Metarules
Rule ext. princpls.
LP rules
Object-grammar
(CF-PS rules)
First, there is a set of basic rules. These are
immediate-dominance-rule (IDR) doubles, ordered pairs (n,i), where n is
the rule number and i an IDR.
IDRs closely resemble CF-PS rules, but, whereas the CF-PS rule
(75a) contains information about both immediate dominance and linear
precedence in the subtree to be accepted, the corresponding IDR (75b) en-
codes only information about immediate dominance.
(75)
a. 'I --> 8
1
8
2
8
n
b. 'I --> 8
1
, 8
2
, , 8
n
The order of the right-hand-side symbols, which are separated in
IDRs by commas, has no significance.
gMore declarative definitions of this procedural model are in preparation (Sag, personal
communication).
The Metagrammar
35
Example (76) could be the basic rule for simple English transitive-
verb phrases:
This rule is equivalent to (77):
Metarules are the second kind of rules. They map IDR doubles to
other IDR doubles. Metarules apply both to basic rules and to rules
derived by applications of metarules. Viewed formally, metarules are
relations between sets of IDRs. They are written in the format A=} B,
where A and B are rule templates. This can be read as: For every IDR
double in the grammar that matches A, the grammar also contains an
IDR double of form B. In each case, the rule number is copied from A to
B.10
Metarule (78) is a fictitious, oversimplified passive rule for English.
(78) V
2
~ V, N
2
, X =} V
2
~ V, X, (p2)
+PASS +by
Variable X ranges over strings of terminal and nonterminal symbols
separated by commas. Variables such as X whose range is not exten-
sionally specified in the grammar, are called essential variables.
The metarule states that, for every ID rule that expands a verb
phrase to a verb, a noun phrase, and some optional additional material,
there is also a ID rule that expands a passive verb phrase to a verb, to
whatever the X ranges over in the input rule, and to an optional preposi-
tional phrase. The prepositional phrase carries a feature -l-by. This fea-
ture, by virtue of some other mechanism, ensures that the optional agen-
tive phrase will be marked by the appropriate preposition.
The application of (78) to (76) or (77) yields (79):
(79) (23; V
2
~ V, (p2))
+PASS +by
In this case, X stands for the empty string. The metarule carries the rule
number over to the output rule. Thus, the verbs that are accepted in
10As was pointed out in Section 3, rule number might be a misleading term for n because
this copying assigns the same integer to the whole class of rules that were derived from
the same basic rules. This rule number propagation is a prerequisite of the GPSG
account of subcategorization.
verb phrases generated (or admitted) by the output rule are the same
ones that are selected by the input rule.
Several metarules can apply in the derivation of a single IDR double.
However, the principle of Finite Closure, defined by Thompson (1982),
allows every metarule to apply only once in the derivational history of an
IDR double. The invocation of this principle avoids the derivation of in-
finite rule sets. If infinite rule sets were allowed, non-CF, non-CS, and
nonrecursive languages could also be generated.
ll
Another component adds semantic translations and instantiates syn-
tactic features. It maps IDR doubles to IDR triples, which are ordered
triples {n;i;t}, where n is a rule number, i an IDR, and t a translation.
The symbols of the resulting IDRs are fully instantiated feature sets (or
structures) and therefore identical to object grammar symbols. These
triples are like the rules of the object grammar except that they do not
indicate constituent order. The mapping is controlled by a set of rule
extension principles including feature instantiation principles, as well
as by an algorithm that assigns the right kind of translation to each rule
on the basis of the encoded syntactic information.
The semantic translation algorithm will be described in Section 5.
Here I will only list some relevant feature instantiation-mechanisms.
Feature cooccurrence restrictions (FCR) are ordered pairs of feature sets
("I, 8), usually written as 'I -+ 8. They are rules that constrain the set of
categories. FCR 'I -+ 8 is interpreted as the statement: if a category in-
cludes 'I, it also includes 8. For example, +MASSN --> +SING states
that a mass noun is singular.
Default value assignments (DVA) specify default values for features.
If a feature has a DVA, the feature has its default value unless otherwise
specified by an ID rule. DVAs might stipulate that the default value for
a feature VOICE is active or that the default for case in English is the
object case.
The Head Feature Convention (HFC) copies all head features (a
designated subset of features) from the mother node onto the syntactic
head of the rule. In some versions of GPSG, unification is used for head
feature percolation. In rule (79), only the mother category had to be
marked +PASS; the feature will be copied onto the verb, since the verb
is the head of the rule and PASS (along with any multivalued features
for voice or verb form) is a head feature.
Some recent unification-based versions of the framework also include
a Foot Feature Convention (FFC). The FFC percolates certain features
called foot features. The values for foot features of a mother category
are unified with its daughters' foot features. If a gap feature is used to
link filler and gap in unbounded dependencies, this feature must per-
colate up to the mother category even if the gap is not in the head of the
rule. Therefore, gap features of this kind are foot features.
llFor a discussion, see Uszkoreit and Peters (1983) and Shieber et al. (1g83a).
The Metagrammar 37
Finally, there is the Control Agreement Principle (CAP) that re-
quires that a controller and a controllee-usually an argument and a
functor category (see Klein and Sag (19S2) for an explicit
definition)-share the values of their agreement features. Agreement fea-
tures constitute another designated feature set. Depending on the version
of the framework, agreement is defined either in terms of identity check-
ing or unification. This principle encodes the fact that agreement always
holds between functor and argument categories. The control relation is
defined in such a way that it always triggers agreement. It therefore be-
comes unnecessary to state agreement feature checks on individual rules.
All features whose values are not determined by any of the rules and
principles will be freely instantiated-i.e., all permitted value assign-
ments will be realized in object grammar rules.
The IDR triple (SO) is an example-though still simplified-of the
result of applying a rule extension component to IDR double (76)
(80) (23; V
2
~ V,
+ACf +ACf +OC
+PL +PL +SG
+3rd +3rd +MASSN
The rule has assigned features for object case, singular number, and the
mass noun property of the noun phrase. The verb phrase -and through
the also the verb-carries features that mark active voice, plural num-
ber, and third person.
The last component of the metagrammar maps the IDR triples to the
rules of the object grammar. For each IDR triple, all the object gram-
mar triples are generated whose CF-PS rules conform to the linear
precedence (LP) rules, the fourth rule set of the metagrammar. LP rules
are members of the LP relation, a partial ordering on VT U VN' An LP
rule (a, 13) is usually written as a < 13 and simply states that a precedes
13 whenever they both occur in the right-hand side of the same CF-PS
rule. Example (SI) might be an appropriate LP rule of English:
(81) V < N
2
The rule will disallow any CF-PS rule in whose right-hand side a noun
phrase category linearly precedes a verb category. Thus, only one CF-PS
rule can be derived from the IDR triple (SO):
(82) (23; V
2
~ V N
2
; V' CN2 '))
+ACf +ACf +OC
+PL +PL +SG
+3rd +3rd +MASSN
It is the separation of linear precedence from immediate-dominance
statements in the metagrammar that is referred to as ID/LP format.
This extension to GPSG, which was first sugested by Gazdar and Pullum
(1982b), makes the theory attractive for application to languages with a
high degree of word-order freedom. Potential applications are discussed
in Section 7.
5 The Semantics of GPSG
It was mentioned in Section 3 that all object grammar rules consist of a
syntactic and a semantic rule. The semantic rules are expressions in in-
tensional logic that specify how the denotations of the daughter nodes
will be combined to yield the denotation of the mother node. Semantic
translations, notated in intensional logic, are model theoretically inter-
preted in the spirit of Montague (1974b). The semantic rule of (82)
states that the denotation of the verb, (which is a function from noun
phrase intensions to functions from noun phrase intensions to truth-
values) is applied to the intension of the denotation of the object noun
phrase. Therefore, the meaning of the verb phrase will be a function
from noun phrase intensions to the denotations of sentences, i.e., truth-
values.
(83) (23; V
2
-+ V,
+ACf +ACf
N
2
;
+OC
+PL +PL +SG
+3rd +3rd +MASSN
At earlier stages in the development of the framework, all ID rules
contained semantic rules. The translation part of an object grammar PS
rule was determined by the semantic component of the basic rule the PS
rule was derived from, as well as by operations applied to the semantic
part of the rule. There was no prohibition against metarules that did
nothing but change the semantics of a rule. Klein and Sag
(1982) hypothesized the existence of an algorithm that could deduce the
translation schema of any ID rule solely on the basis of the encoded syn-
tactic information. They constructed such an algorithm for a limited
grammar of English. For most ID rules, the algorithm relies completely
on the assignment of semantic types to syntactic categories. The default
combination operation is functional application, as in (83).
However, there are cases in which either other semantic operations
have to be used or additional syntactic information must be evaluated.
For raising and equi verbs, straightforward functional application does
not allow the appropriate binding of the complement's subject. Since in
GPSG syntax no subject node is created for infinitival complements, ad-
ditional semantic operations are necessary for this binding. For formula-
tions of such operations, see Klein and Sag (1982).
The Semantics of GPSG 39
The algorithm needs further syntactic information if a functor can be
applied to arguments in more than a single order, but only one order is
the correct one. An example is English ditransitive verb phrases. The
algorithm cannot determine the order in which the verb applies to t.he
objects by considering semantic types alone, since both objects are of t.he
same type. Since standard versions of GPSG do not encode grammatical
functions in their categorial information, there is no way to specify t.he
function-argument structure in the syntactic component of the rule. In
English, the two objects are distinguished only by their position in t.he
sentence. Therefore, semantics needs access to their linear order.
However, in the procedural model of the metagrammar, the LP com-
ponent is the last metagrammatical device to apply in the derivation of a
rule. The first solution to the problem (Klein and Sag (1982)) involves a
type of LP rule that is allowed to look inside the semantic translation
scheme. An LP rule of this kind can put the two objects in the right
linear order, depending on which of the two combines with the verb first
semantically.
A later proposal (Sag and Klein (1982)), based on a more declarative
interpretation of the metagrammar, allows for the integration of
metagrammatical components. The details of formalization have not
been worked out yet. If this enterprise proves successful, semantic trans-
lation can be made sensitive to the result of LP rule applications. In
current versions of GPSG it is therefore assumed that the metagram-
matical components of Rule Extension Principles and LP rules are col-
lapsed into one component. This integration will come up more than
once in the discussion of German word order. Another related issue that
arises is whether grammatical functions are nevertheless needed for the
mapping from syntactic categories to semantic function-argument posi-
tions.
6 Notes on the Generative Power of GPSG
In the introduction to this chapter it was mentioned that the question of
generative capacity has been extremely relevant to all research on and
within GSPG. In the course of this book, I will occasionally relate lin-
guistic problems or solutions to formal properties of the grammar. In
this Section I will introduce those questions that are relevant to the ex-
ploration of the formal properties of GPSG.
A generative grammar may be seen as a (potentially) recursive
characterization of a set of strings over an alphabet. This set shows t.he
weak generative capacity of the grammar. If a grammar generates (or
recognizes) a string, it does so by assigning the string a structure. The
set of all ordered pairs whose first member is a string accepted or
generated by the grammar and whose second member is a structure as-
signed to this string by the grammar indicates the strong generative
capacity of a grammar .12 Weak and strong generative capacity of a
framework, i.e., of a class of grammars, depends on the weak and the
strong generative capacities of all the individual grammars.
There is a long tradition of research directed toward determining and
formalizing the generative capacity (especially the weak kind) of gram-
matical frameworks. What are the reasons for this type of research?
Linguistic frameworks are usually an essential part of a synchronic
theory of natural language. A theory that rules out grammars that could
not possibly be grammars of human languages is obviously superior to
one that lacks this property. Unfortunately, not enough is known about
the formal properties of natural languages to allow this kind of evalua-
tion without a number of hypothetical stipulations.
The question is which formal properties of languages are likely to
apply to natural languages. There are a number of observations that are
relevant for hypotheses of this kind. The most important are as follows:
Natural languages are nonfinite.
They can be learned by a child in a (brief) finite time after
exposure to a finite set of samples.
They can be understood in real time, i.e., almost in the time
it takes to utter them.
Natural languages also resemble one another in a number of ways:
They are learned in about the same length of time.
Their semantics are closely related.
They are governed by the same articulatory and auditory
constraints.
They have a similar degree of information density.
The problem is to determine how the class of all possible languages
can be subdivided into formally definable subclasses, with known
properties as to their parsing time, decidability, learnability, etc. One of
the earliest and probably the most influential subdivision model is known
as the Chomsky Hierarchy. The classification is based on different types
of rewriting systems that can serve as generative grammars. Each type
of rewriting system specifies the class of languages that can be generated
by such grammars.
For every two subsets A and B, either A is properly included in B or
B is properly included in A. The Chomsky Hierarchy is best known as
the corresponding Venn Diagram with its concentric circles:
12What qualifies a structure can radically, differ depending on the type of grammar. It
might include representations that do not at all resemble phrase structure trees.
Notes on the Generative Power of GPSG 41
It turns out that each class of languages also corresponds to an automata
type that recognizes them.
For a while, research in generative capacity concentrated on two
questions: What is the place of the weak generative capacity of a given
framework in the Chomsky hierarchy? Where do the natural languages
that we know fit?
Some particularly interesting results have been obtained, some less
so. Probably the most influential result is the chain of research set off
by Peters and Ritchie (1973) who were able to prove that unconstrained
transformational grammars have Turing machine power, i.e., that they
can generate any recursively enumerable set.
The origin and development of GPSG were very much influenced by
these findings, for none of the many attempts to constrain TG had led to
a satisfactory result. In other words, none of them constrained the
generative power sufficiently by linguistically motivated or adequate
restrictions. GPSG is built on the assumption that natural languages are
context-free. From the very beginning, one of the main arguments in
support of GPSG has been that the framework is constrained in its
generative power. Gazdar and Pullum (1982d) have refuted the COLn-
terexamples that are most commonly employed to disprove the context-
freeness of natural language.
The problem of assuring that GPSG will be context-free is not as
easy and straightforward as originally supposed. It soon became clear
that the essential variables in metarules made the metagrammar power-
ful enough to create infinite rule sets, moreover infinite rule sets that
generated non-context-free languages. A conjecture by Joshi1
3
stating
that the restriction to at most one essential variable per metarule would
keep the framework context-free was refuted by Uszkoreit and Peters
(1983). This result by itself is not necessarily damaging, since we do not
know for sure whether natural languages are context-free. However,
Uszkoreit and Peters also proved that metarules with not more than one
essential variable can generate grammars for all recursively enumerable
languages. A framework with Turing machine power, however, is clearly
too unconstrained.
14
A number of constraints on metarules have been proposed. For a
detailed discussion of most of these constraints see Shieber, Stucky,
Uszkoreit, and Robinson (1983a)(SSUR). The bottom line of this study is
that none of them is both linguistically adequate and sufficient in its ef-
fects on the power of the formalism.
There are two potential solutions to the problem that merit closer
attention. Both of them are relevant here because they can be evaluated
on the basis of data about word order.
One strategy has been adapted in the standard versions of GPSG un-
der the name of Finite Closure. It was formalized by Henry Thompson
(1982). The constraint states that every metarule can apply at most once
in the derivation of any given object grammar rule. Finite Closure en-
sures that the object grammar will be finite and therefore also context-
free, provided that infinite rule sets do not arise through some other
metagrammatical device. SSUR cite counterexamples to the constraint
from Japanese, German, and Warlpiri. The question of whether the con-
straint of Finite Closure can be imposed on a GPSG for German will be
discussed in Chapter 6, Section 3.2.
The second strategy, suggested by SSUR as a potential remedy for
the capacity problem, is the elimination of the metarule component.
This solution would be in the spirit of Gazdar (1982a) (p.132) who, refer-
ring to the transformational component, remarks: II The strongest way to
constrain a component is to eliminate it. II
13Gerald Gazdar gives Tom Wasow the credit for having seen the problem first, Wasow's
observation, as well as Joshi's conjecture, is reported by Gazdar (1982a)(footnote 28).
14It needs to be emphasized that little is known about the formal properties of current
versions of GPSG. Uszkoreit and Peters (1983) have pointed out that the intricate
interaction of the numerous metagrammatical devices can result in nontrivial effects on
the generative power of the formalism. Their results are restricted to a general version of
metarule phrase-structure grammars, as it is defined in the appendix to their paper.
Notes on the Generative Power of GPSG 43
As radical as this suggestion might seem at first glance, a closer look
at the development of the metagrammatical component of GPSG shows
an increasing degree of division of labor. Many classes of phenomena
that were handled by metarules in the past are now assigned to other
metagrammatical devices, such as LP rules or feature instantation prin-
ciples. A recently proposed restriction on metarules can be viewed as an
indication of this specialization process. First introduced by Daniel
Flickinger (1983), it is called the Lexical Head Constraint. It restricts
the application of metarules to those rules that have a lexical head. The
underlying idea is that metarules should affect only syntactic functors
and their arguments. If it proves adequate, metarules might be viewed
as lexical rules, independent of whether they are formally part of the syn-
tax or of the lexicon. Their power could be greatly reduced, since there
is only a small number of finite syntactic arguments for every functor.
The constraint would correspond to the lexical treatment of
relation-changing rules in LFG (Bresnan (1982)).
However, there is one type of metarule that defies this constraint and
proscribes the complete elimination of the metarule component. These
metarules II flatten II the constituent structure. They are a prerequisite of
phrase structure solutions to many word order problems. Metarules of
this kind are called liberation rules (Pullum (1982)).
At this time, both suggestions for solving the problem of excessive
generative power are research strategies at best. More empirical data are
urgently needed. This book will contribute pertinent data from German
to further research on the problem.
Some remarks on the direction of this research are in order. During
the last couple of years general opinion as to the significance of genera-
tive capacity results has become more sophisticated. An indication of t.he
change in attitude is Berwick's and Weinberg's criticism (1982) of ove:rly
naive approaches. The traditional classes of languages placed in t.he
Chomsky Hierarchy are no longer regarded as the appropriate measure;
additional classifications are under investigation, such as indexed
languages or Joshi's (1983) tree-adjoining languages. Other properties
of natural languages have been investigated such as: profligate larujuaqes
or constant growth (Pullum (1983), Joshi (1983)).
At about the same time, the investigation of the crossing depen-
dencies in Dutch verb phrases (Bresnan et al. (1982)) has provided a
powerful argument against the notion of a strongly context-free syntax.
Research in progress on related phenomena in Swiss German might even
extend this argument to apply to weakly context-free grammars.J'' Some
proponents of GPSG no longer insist that the framework has to remain
context-free. However, the Finite Closure constraint is still widely used
15Since this was first written, Shieber (1984) has has finished his research. His evidence
suggests that Swiss German is not context-free.
to ensure that current versions will actually have the property of context-
freeness.
This constraint is apt to be too inflexible to allow for mildly
context-sensitive languages. If one sets an upper bound on the number
of applications of a metarule in any rule derivation, then the object
grammar will be context-free. If one does not stipulate such a number,
metarules increase the generative power too much.
This book is not concerned with the uncovering of new revelations
about the generative power of the GPSG formalism. However, the con-
sideration of formal properties is important for the GPSG analysis of
German word order. It will be shown that the grammar for a fragment
of the language, presented in Chapter 3, does not require an increase in
formal power. The question will come up again in the discussion of cer-
tain phenomena, such as free adjuncts, extraposition, and topicalization
in Chapter 6. The possibility of eliminating the metarule component will
be relevant in the investigation of the interaction between metarules and
the ID/LP mechanism. There is another connection between the formal
power of the framework and my analysis of German: in Chapter 6 I will
discuss briefly the implementability of the analysis, including the
modifications of the framework.
7 GPSG and Word Order Freedom
The framework makes a number of predictions about the degrees of word
order freedom that one can expect to find in human language. It also
presupposes and specifies a close relationship between ordering freedom
and constituent structure. Real permutational variation is restricted to
sibling constituents. All kinds of ordering variability that permeate
traditionally assumed constituent boundaries must either be handled by
category-valued feature percolation or require modifications in the im-
posed constituent structure. Category-valued feature percolation is
restricted to long-distance dependencies. There is only a small set of
designated category-valued features. Each of these features corresponds
to a certain syntactic process, such as Wh-Movement or Relative-Clause
E
.. 16
xtraposition.
One modification of the constituent structure that can be made to
accommodate scrambling across traditional constituent boundaries is a
flattening of structure. This is not just a technical trick; the imposed
constituent structure has to fit into the overall syntax and semantics of
the language. Therefore, the GPSG approach to constituent scrambling
neither introduces nor abolishes criteria for constituenthood. It might
merely assign different weights to some of the existing criteria, such as
16See Maling and Zaenen (19S0) for the encoding of multiple extraction processes,
GPSG and Word Order Freedom 45
distributional properties, semantics, agreement, government, crO:3S-
linguistic generalizations, historical considerations, and phonological
evidence.
When Pullum (1982) investigates potential applications of the ID/LP
version of GPSG to languages with free word order phenomena, he also
discusses a type of metarule that has the effect of creating flatter con-
stituent structures. He calls these rules (already widely used before that)
liberation rules. They have the following form:
(85) Q' -+ <J>, f3 =} Q' -+ <J>, IJ!
The idea behind liberation rules is simple and intriguing.
Scrambling or other reordering phenomena often suggest a flat structure,
whereas other generalizations could be stated more easily by assuming; a
deeper structure. An example used in Gazdar (1982a) is the verb phrase
in VSO languages. In some VSO languages, such as Breton (Anderson
and Chung (1977), pp. 22-24) and Modern Irish (McCloskey (1979)), verb
phrases can be found in certain constructions, whereas there is no indica-
tion of a VP constituent in the majority of sentences. An example cited
by Pullum are noun phrases in languages that allow for scrambling out
of NPs, such as certain Australian languages. The following metarule
might be assumed:
(86) S -+ X, NP =} S -+ X, ADJ, N
Now ADJ and N can be separated from each other when the rule is
linearized in accordance with the LP rules.
There are certain discomforting consequences connected with the use
of liberation rules. As the name liberation rule suggests, the string that
replaces a category symbol in the output rule represents a possible expan-
sion of this category. Metarule (86) II liberates II the constituents that
would otherwise be dominated by a NP. Thus, every sentence in wh.ch
ADJ and N are adjacent to each other and in which their order does not
violate an LP rule will be ambiguous.
Another consequence is more serious. One of the main reasons be-
hind the suggestion of metarules such as (86) is the desire to encode syn-
tactic relations by relating rules that generate similar structures through
the metagrammar. However, one important relationship cannot be ex-
pressed without extending the framework; this is the relationship between
17To be precise, if> and I/t range over strings of terminal and nonterminal symbols that are
separated by commas.
the ID rule that "normally" expands NP to ADJ and N, on the one
hand, and the output rule of the metarule, on the other. In fact, this
rule could also have served as the input rule for metarule (86).IS
A third but closely connected disadvantage follows directly from the
inability to relate the output rule to two input rules. Let us assume that
the language has several rules expanding noun phrases and, furthermore,
that it is a general phenomenon that material can be scrambled out of a
noun phrase. Then for each of the other NP rules a metarule like (86)
needs to be written that liberate its right-hand-side consituents in a cor-
responding clause rule. The solution is obvious. Liberation rules should
be considered metarules that map two input rules to an output rule:
(87) Q' -+ <J>, (3
(3-+1J!
Now we can write the following, more general formulation of (86),
using two essential string variables, X and Y:
(88) S -+ X, NP
=} S -+ X, y
NP -+ y
Although so-called double-barrelled metarules have been considered in in-
formal discussions, the problems connected with extending the framework
formally in this direction have not yet been seriously investigated.
In Section 6 the option was discussed to eliminate the metarule com-
ponent altogether. As SSUR (1983a) have pointed out, liberation rules
are one of the few rule types that cannot yet be replaced by other
metagrammatical devices. Their special properties seem to indicate that
they could be singled out and transformed into a separate formalism.
Not only will metarules that liberate constituents play a role in the
investigation of German word order, but also their
counterparts-metarules that combine constituents. I will not give ex-
amples at this point, but merely mention the possibility of such rules.
Just assume the reverse of metarule scheme (87):
(89) Q' -+ <J>, (3
Q' -+ <J>, IJ!
IS An example of a metarule that has a rule expanding the liberated constituent as its input
rule is presented in Gazdar, Pullum, and Sag (19S2a). The metarule, expressed in pre
ID/LP notation, inserts an adverb after the first auxiliary verb.
GPSG and Word Order Freedom 47
Although metarules of this kind will not be proposed in the grammar for
the fragment, they will be relevant in the discussion of German
topicalization.
Chapter 3
The Order of Main and Auxiliary Verbs
1 Introduction
The discussion of the basic problems of German word order, and the
brief outline of the current GPSG framework have set the stage for the
first step in the presentation of a phrase structure analysis of a fragment
of Standard German. Frequently, this grammar will be referred to as
GFG.
The fragment includes simple intransitive, transitive, and ditransitive
sentences with some auxiliaries and modals, and with separable prefix
verbs. The fragment does not include questions or imperatives-mainly
because any discussion of their semantics is outside the scope of this
book. However, for both imperatives and questions, the presented gram-
mar generates the clause types that exhibit the appropriate word order.
Negation is also excluded from the selected framework. I do not analyze
the internal structure of noun phrases, since, in my discussion of German
word order, nothing will depend on this. The analysis of the auxiliary
system is restricted to those aspects that are of theoretical interest in
connection with the topic of the book.
This chapter is concerned with the position of verbs in the sentences
of the fragment. The notations I will use follow the ones introduced in
Chapter 3 and agree for the most part with the conventions that have
been established in standard publications of GPSG. A version of
X-syntax is assumed that has three bar levels. Bar levels are indicated
by superscripts. The maximal projections of A, N, Pare A
2
, N
2
, p2; the
maximal projection of V is V
3
, the category of sentences. Categories in
rules are usually underspecified: only those features are mentioned that
are relevant for the current analysis of German. For expository purpose
I will for the most part restrict myself to the use of binary features. Any
binary feature system can easily be transformed into a multivalued sys-
tem.
48
Introduction 49
The presentation of the analysis starts with some basic rules and
metarules that generate a grammar for deriving simple verb-initial and
verb-final sentences. Next the fragment will be extended to include
auxiliaries and modals. Then an analysis of the first position in verb-
second clauses will be presented that regards all first-position con-
stituents as fillers for long-distance extractions.
2 The Position of the Main Verb
2.1 Basic VP Rules and Flat Clauses
One of the central issues in research on languages with word order
freedom is the flatness of the phrase structure. I will postpone any
theoretical discussion of the issue until I have examined a number of
specific instances in which a choice needs to be made between a flat
structure and a more hierarchical one, This is probably best exemplified
by the debate over the existence of verb phrases in languages (e.g., typi-
cal VSO languages) in which they could not be deduced from distribu-
tional facts. Although German verb phrases are not always
discontinuous, the debate does extend to German. Sentence (90) shows
the same constituent order as its English translation (91).
(90) Der Doktor gibt dem Patienten die Pille.
(91) The doctor gives the patient the pill.
One could easily impose on (90) the constituent structure (92)
generally assumed for (91).
The doctor gives the patient the pill
If the same constituent structure is also imposed on the five grammatical
permutations of (90), (93a)-(93e), two of them will contain discontinuous
verb phrases, i.e., (93c) and (93e).
50 The Order of Main and Auxiliary Verbs
(93)
a. Der Doktor gibt die Pille dem Patienten.
b. Die Pille gibt der Doktor dem Patienten.
c. Dem Patienten gibt der Doktor die Pille.
d. Dem Patienten gibt die Pille der Doktor.
e. Die Pille gibt dem Patienten der Doktor.
If a noun phrase-verb phrase clause structure is assigned to a verb-
second clause with an auxiliary such as (94), only two permutations will
produce a continuous verb phrase-i.e., the two that have the subject in
first position.
(94) Der Doktor wird dem Patienten die Pille geben.
In all permutations of the verb-initial counterpart of (94), the verb
phrase forms a discontinous constituent. Only subject-initial clauses can
be regarded as clear cases of NP-VP surface structure.
However, an additional consideration is the high frequency of sen-
tences that are not introduced by the subject. Jecklin (1973), in his
statistical investigation of clause constructions in spoken language, finds
that fewer than 60% of the clauses in his corpora start with the subject.
It can be expected that the percentage of sentences that do not begin that
way is even higher in written language, where intonation cannot be used
to mark the discourse role of the phrases.
I have not yet found any convincing evidence for the existence of
English-like VP nodes in German clause structures.t'' That is not to say
that German is lacking verb phrases. There are a number of construc-
tions that have to be classified by any distributional account as verb
phrases. Example (95) shows a participle group in adjunct position, (96)
has the same participle group in prenominal position, and (97) contains a
. fi .. I I 20
to-m imtrva comp ement.
(95) Peter lief umher, das verlorene Buch suchend.
Peter ran around the lost book searching
Peter ran around searching for the lost book.
19There are, however, several apparent subject-object asymmetries that have been used to
argue against a flat clause structure. They are discussed in the next subsection,
20Although a clausal analysis of these infinitival complements and adjuncts, as it is as-
sumed in versions of transformational grammars and other multilevel frameworks, would
not be impossible for the surface-structure-oriented framework of GPSG, it would
nevertheless violate the spirit of the theory. The only filler-gap dependencies in GPSG
are long-distance dependencies.
The Position of the Main Verb sr
(96) Der das verlorene Buch suchende Mann ist Peter.
the the lost book searching man is Peter
The man searching for the lost book is Peter.
(97) Peter versucht, das verlorene Buch zu finden.
Peter tries the lost book to find
Peter is trying to find the lost book.
Clearly, for multilevel frameworks the apparent contradiction is not
a problem. One level of the syntactic representation could always con-
tain the hierarchical structure-in this case the VP-but for clauses
there might be another level that does not possess a VP node.
An un enriched phrase-structure grammar could also try to account
for both flat clause structures and VP constituents, but not without
losing some generalization. If a clause is generated as a flat structure
without a VP node but if there are also VP rules, then there has to be
one VP rule and one corresponding S rule for every subcategorization
frame.
It is not the existence of closely related rules that is bothersome but
the inability to express this relation in some nonredundant way. This is
where a metagrammar is especially convenient. Rules (98a)-(98d) are
basic rules of the metagrammar for the fragment. Each rule is followed
by some examples for verbs that V can expand to.
(98)
a. (5, y2 -+ y)
kommen, laufen, schlafen
(come) (run,go) (sleep)
b. (6, y2 -+ v, N
2
) kennen, suchen, nehmen
+ACC (know) (search) (take)
c. (7, y2 -+ v, N
2
) helfen, begegnen, vertrauen
+DAT (help) (meet) (trust)
(2 2 2)
geben, schicken, zeigen d. 8, Y -+ Y, N , N
+ACC +DAT (give) (send) (show)
Keep in mind that basic rules are ID rules; the commas are supposed
to indicate this. The noun phrases that function as the direct and in-
direct objects are required to have accusative and dative case, respec-
tively.
The following metarule generates the corresponding flat clause rules:
(99) y2 -+ X =} y3 -+ N
2
, X
+NOM
The metarule simply states that, for every VP rule that expands to some
string of symbols, there is also a rule expanding S to the same string of
symbols, plus an additional NP in the nominative case. The right-hand-
sides of the resulting S rules are also unordered.
At first glance it may seem that a very intricate mechanism will be
needed to encode subject-verb agreement. The reason for this is that the
finite verb that has to agree with the subject in person and number may
not be the sibling of the subject. Only if the main verb is the finite verb
of the clause, subject and verb are siblings. The problem could be for-
mulated as follows: if the main verb is finite then the subject needs to
agree with it, if not, then the agreement features of the noun need to be
percolated up to provide for an agreement with a finite auxiliary or
modal higher up in the tree. Such a conditional agreement statement
would not only destroy the beauty of a percolation and feature agree-
ment system, driven by very general principles such as the Head Feature
Convention and the Control Agreement Principle, it would also result in
contradictory statements.
The agreement features are head features. This allows them to
travel up the tree from head to head and enforce agreement on many
levels such as in the complex NP in (100).
(100) N
2
De~~l
I .r >.
dem A N
1
I <>.
kleinen A N
1
I I
griinen N
I
Mann
All constituents in this tree have to agree in case and number.
However, if PERSON and NUMBER are head features, then the values
for those features on the clause node will always come from the main
verb. Therefore the agreement features from the noun could not be per-
colated up unless they agree with the main verb in which case it is of no
importance to have them travel up the tree.
The solution to this and many similar problems lies in modern
unification based feature mechanisms as they have been developed for
UG, LFG, PATR, and GPSG.
21
Redefinitions of agreement in terms of
21Kay (1979), Bresnan and Kaplan (1982), Shieber et al. (1983b), Gazdar and Pullum
(1982a).
The Position of the Main Verb 53
unification permit underspecified agreement features such as an empty
number feature for the English determiners the, a and no. They can also
allow for an agreement of subject and verb in German flat clauses. As-
sume that PERSON and NUMBER are unspecified for nonfinite verb
forms. This is at least as intuitive as a minus value in a binary feature
system or a NIL or t. value in multivalued systems. If agreement is
defined as unification, then the subject will always agree with nonfinite
verb forms. The result of the unification is unified with the agreement
features of the mother node. Note, that it does not matter in which se-
quence the unifications are performed and from which node-subject or
verb-the values for the mother nodes' features come, since unification is
associative. Unification can only fail if a subject has PERSON and
NUMBER values that differ from the main verb. This in turn can only
happen, if the main verb is finite. In Section 3, I shall return to the
question of subject-verb agreement and show how the agreement works
for sentences with auxiliaries or modals.
Since subject-verb agreement is the only place in the grammar for
the fragment in which a unification based feature system needs to be in-
voked for demonstrating the basics of the rule system, I will not unnec-
cessarily complicate the exposition of the remaining rules by adapting my
notation to a more sophisticated feature mechanism. The only modifica-
tion that needed to be made in the formulation of metarule (99) is the
agreement between clause and subject noun-phrase, indicated in (101) by
the notational variable Q' for PERSON and NUMBER.
(101) V
2
-+ X =} y3 -+ N
2
, X
Q' +NOM
2.2 Apparent Subject-Object Asymmetries
One class of potential arguments against a flat clause structure without a
VP node builds on apparent differences in syntactic behavior that ho..d
between subjects and objects. In the proposed grammar for the fragment
there are only two properties that distinguish subjects from other ar-
guments of the verb: 1. subjects appear only in clause rules: 2. subjects
agree with the finite verb in person and number. However, subjects are
neither marked as such, nor are they recognizable by their position in the
phrase structure of a sentence. It could therefore well be the case that
the analysis in GFG could not account for additional differences between
the subject on one side and other verb arguments on the other. For-
tunately, the subject-object asymmetries cited in the literature turn out
to be irrelevant for the status of the subject. The easiest way to support
this claim is to look at the observations one by one.
The first apparent asymmetry to be considered is cited by Lenerz
(1977). As it was pointed out in Section 5, the first position in verb-
second clauses is occupied either by the topic of the sentence or by a con-
stituent that represents an emphatic or a contrastive focus. It has been
observed, however, that in those cases in which the first position con-
stituent is a personal pronoun, it has to represent a contrastive or em-
phatic focus unless it is the subject. Example (102) seems only accept-
able if ihn carries emphatic or contrastive stress.
(102) Ihn sehe ich oft.
him see I often
Him, I see often.
If this observation were true as formulated above, it would yield a
generalization about a difference between subjects and objects.
22
However, consider the following examples.
(103) Dann besuchte ich Paul.
then visited I Paul
Ihm war gerade ein
him was just a
Pfandungsbescheid zugeschickt worden.
distress-warrant sent was
Then I visited Paul. He had just been sent a distress-warrant.
(104) Ihn fror.
him cold-was
He was cold.
(105) Ihm graute es vor den Folgen seines
him horrified it before the consequences of-his
Bankrotts.
bankruptcy
He was horrified by the consequences of his bankruptcy.
(106) Ihm war dieses Jahr wirklich alles miBlungen.
him was this year really everything wrong-went
Really everything had gone wrong for him this year.
None of the personal pronouns in initial position have to be inter-
preted as a contrastive or emphatic focus. The initial observation needs
22This does not imply that GFG could not easily incorporate the generalization. In fact, it
could. Yet the generalization might still contribute to an argument for an analysis that
assigns the subject a more distinguished role than GFG does.
The Position of the Main Verb 55
to be modified. It seems that certain indirect and agentive objects do
not share the observed behavior of objects.
The next asymmetry to be discussed was cited by Craig Thiersch
(1982). In general, the nonfinite main verb can be fronted together with
arguments, adjuncts, or auxiliary verbs.
23
The original claim was that
subjects could not be fronted together with a verb. This difference
would explain the inacceptability of (107a) in contrast to the acceptable
sentence (107b).
(107)
a. *Der Vater zeigen wollte dem Kind das Buch.
the father show wanted the child the book
b. Das Buch zeigen wollte der Vater dem Kind.
the book show wanted the father the child
The father wanted to show the child the book.
However, Craig Thiersch later became aware of cases in which the
subject can be fronted with a verb.
24
Consider the following sentence
with a nonagentive subject of a so-called psychological verb.
(108) Solch ein Fehler unterlaufen war ihm noch nie.
such a mistake undergone was him still never
Until now, he had never made such a mistake.
A closer investigation of the phenomena yields the following picture.
Certain adverbial phrases such as local, temporal, and directional
phrases, can accompany a fronted verb most easily. Next in the hierar-
chy are direct objects. Some authors such as Heidolph et al. (1981) and
Nerbonne (1982a) do not consider the fronting of verbs together with an
indirect object grammatical. Yet, all informants who were asked in this
investigation accepted at least some of these frontings. Finally, subjects
can only be fronted with a verb if they are nonagentive subjects of cer-
tain verbs that take agentive objects.
Another relevant observation concerns discontinuous noun phrases
with the complex inerrogative determiner was fur ein. In widespread
dialects of Standard German, sentences such as (109) are acceptable,
whereas the corresponding sentences of type (110) are considered ungram-
matical.
23These constructions are not included in the fragment, For a discussion see Chapter 6,
Section 3,5.
24personal communication
(109) Was hat Paul fur Leute getroffen?
what has Paul for people met
What kind of people did Paul meet?
(110) *Was haben uns fur Leute geholfen?
what have us for people helped
The strong contrast disappears for many speakers if we substitute
the questions by emphatic interjections containing the particle da.
(111) Was hat Paul denn da fur Leute getroffen!
what has Paul then there for people met
What kind of people did Paul ever meet there!
(112) Was haben uns denn da fur Leute geholfen!
what have us then there for people helped
What kind of people ever helped us there!
However, Irene Heim
25
has observed that the subject of psychological
verbs can be realized as discontinuous was fur ein NPs even where other
subjects cannot.
(113) Was sind Paul fur Fehler unterlaufen.
what are Paul for mistakes undergone
What mistakes did Paul make.
Sentences like (114) are generally much more readily accepted than
their counterparts represented by (115).
(114) Was hast du den Leuten fur BUcher verkauft?
what have you the people for books sold
What books did you sell the people?
(115) Was hast du fur Leuten deine BUcher verkauft?
what have you for people your books sold
What people did you sell your books?
Again, we have a hierarchy with the order direct object, indirect ob-
ject, subject.
The last observation to be considered in this context concerns the ex-
tractability of phrases from tensed subordinate clauses. The traditional
linguistic wisdom about long distance dependencies in Standard German
25 I ..
persona communication
The Position of the Main Verb
[j7
has been that they are clause bound. Under any interpretation of the
term, this means that they cannot reach into tensed subclauses.
However, there is a growing tendency in German to extract from tensed
clauses under certain conditions. Hans Boas
26
has pointed out that there
is a hierarchy of accessibility, which he compared with Keenan and
Comrie's accessibility hierarchy (1977). The most frequent example of
extractions from tensed clauses involve adverbial phrases-especially
temporals, locals, or directionals.
(116) In welchem Jahr glaubst du,
in which year believe you
daB ich sie zum letzten Mal gesehen habe?
that I her to-the last time seen have
In which year do you think that I saw her for the last time?
Less frequent, and for many speakers less acceptable, are extractions
of objects.
(117) Wen glaubst du, daB ich gestern gesehen habe?
whom believe you that I yesterday seen have
lhom do you think that I saw yesterday?
Extractions of subjects such as in (118) are limited to very restricted
dialects of Standard German that are probably influenced by regional
dialects.
(118) Wer glaubst du, daB den Mann gesehen hat?
who believe you that the man seen has
lho do you think saw the man?
Again, there seems to be a real contrast between subjects and ob-
jects. And again subjects of psychological verbs behave more like ob-
jects.
(119)
a. Welche Aufgaben glaubst du, daB ihm am leichtesten
which tasks believe you that him at easiest
fallen?
fall
lhich tasks do you think he can handle most easily?
26Talk at The University of Texas in July 1979.
58
b. Welche Gerichte glaubst du, daB ihm am besten
which recipies believe you that him at best
gelingen?
[Insert]
Which recipes do you think come out best for him?
If some speakers consider these examples still less acceptable than the
object extractions, then this might be due to another regularity, which
can be described here only very briefly.
Consider the following pairs of sentences.
(120)
a. Wann glaubst du daB du den Mann gesehen hast.
when believe you that you the man seen have
When did you think you saw the man?
b. Wann glaubst du am Abend daB du den Mann gesehen
when believe you at evening that you the man seen
hast?
have
When in the evening did you think you saw the man?
(121)
a. Wen hattest du gesagt daB wir anrufen solIten?
whom had you said that we call should
Whom had you said that we should call?
b. Wen hattest du den Mann gebeten daB er anrufen
whom had you the man asked that he call
sollte.
should
Whom had you asked the man to call?
Clearly, whenever the matrix clause contains a constituent of the
kind (case marking, preposition, thematic role, etc.) the extracted con-
stituent belongs to, the sentence sounds worse. Yet, all matrix clauses
whose main verbs subcategorize for a tensed clausal complement contain
a subject. The NPs in (119) are chosen with this fact in mind. The
extracted subjects differ from the matrix subjects in number and
animacy. However, there still remains a slight contamination by their
agreement in case.
Some dialects of German allow for subject cliticization. Not surpris-
ingly, the sentences in (122) with a cliticized matrix subject sound more
acceptable than their counterparts in (119)
59
(122)
a. Welche Aufgaben glaubst'n, daB ihm am leichtesten
which tasks believe you that him at easiest
fallen?
are
Which tasks do you think he can handle most easily?
b. Welche Gerichte glaubst'n, daB ihm am besten
which recipies believe you that him at best
gelingen?
work-out
Which recipies do you think come out best for him?
Differences between the extraction of direct and indirect objects are
minute. However, it seems that the extractions of direct objects come
out a little better.
(123)
a. Was denkst du, daB ich ihm gegeben habe.
what think you that I him given have
What do you think did I give to him?
b. Wem denkst du, daB ich es gegeben habe.
whom think you that I it given have
Whom do you think I gave it to?
There are two striking similarities that connect the four independent
phenomena: 1. in each case the claimed borderline between subjects and
objects is crossed by the arguments of certain verbs-usually verbs with
nonagentive subjects and agentive objects; 2. in all four cases, there ex.st
also contrasts among the different classes of nonsubjects as to their be-
havior with respect to the assumed subject-object asymmetry.
The following conclusions can be drawn from these similarities. One
and the same hierarchy plays a role over and over: subject-indirect
object-direct object-temporal, local, or directional adverbial phrase. This
is the same hierarchy that underlies the unmarked order of verb ar-
guments. The behavior of the arguments of psychological verbs suggests,
however, that the hierarchy should in fact be stated in terms of thematic
roles such as agent and theme instead in terms of grammatical functions.
The high degree of correlation between thematic roles and grammatic al
functions might explain why regularities that obey the hierarchy could be
interpreted as subject-object asymmetries.
These conclusions show a surprising extent of parallelism with
similar conclusions in the discussion of the unmarked order principle
Chapter 6, Section 3.1.1. In both contexts the question arises whether it
would be advantageous to salvage the formulation of the relevant
generalizations in terms of grammatical functions by constructing special
explanations for the exceptions.
27
In Chapter 6, Section 3.1.1, this ques-
tion is raised, debated, and answered negatively for work in the
framework of GPSG.
2.3 Features and Rules
Let me introduce some features of verbs, verb phrases, and sentences.
Verbs can be FIN, that is finite or nonfinite. They can also be marked
AC and MC. MC is the acronym for main clause, AC for assertion
clause. Not too much importance should be assigned to the feature
names; it will become evident later that +MC, for instance, marks
clauses that are like main clauses in form, but do not necessarily always
function as such.
The three verb features can be related to one another by feature
cooccurrence restrictions (FCR) (124a) and (124b).
(124)
a. +AC -+ +MC
b. +MC -+ +FIN
FCR (124a) states that every constituent carrying the feature +AC
also carries the feature +MC. By FCR (124b), +MC constituents will
always be finite; therefore, every main clause must be finite.
The first LP rules I wish to introduce will determine the position of
the verb.
28
(125)
a. V < X
+MC
b. X < V
-MC
27 This general strategy is pursued by relational grammarians who use the Unaccusative
Hypothesis for this purpose.
28Although rules like (125a) and (125b) have been called LP rules in the GPSG work of
which I am aware, they should be referred to as LP rule schemata that translate into
large sets of LP rules, The reason is simple: LP is defined as a relation on categories and
a complex symbol such as V[+MCj abbreviates the full set of categories that are verbs
and possess this feature. Instead of X, I could have written any feature that is shared by
all other categories or at least by those that occur in conjunction with verbs-if such a
feature exists. The redefinition of LP in Chapter 5, Section 2 will simplify things
considerably. It will permit us to regard (125a) and (125b) as LP rules. I will therefore
continue to call them LP rules.
The Position of the Main Verb 1>1
The variable X stands for any terminal or non terminal symbol of the
grammar. The effect of these LP rules is to prevent the generation of all
object grammar rules in which a +MC verb is preceded by any other
constituent or in which a -MC verb is followed by any other constituent.
The tiny metagrammar has now grown to the point at which it can
be put to work to generate the rules for simple German sentences. If one
II turns the crank" long enough, the metagrammar will-in addition to
many others-produce the following two rules:
(126)
a. 6, V3 -+ V N
2
N
2
+MC +NOM +ACC
+FIN
b. 7, V3 -+ N
2
N
2
V
-MC +DAT +NOM
+FIN
Rule (126a) is derived from (98b), (126b) from (98c). The features
MC and are instantiated by the Rule Extension Principle component
in accordance with the FCRs (124a) and (124b). Any metarule applica-
tion copies the rule number onto the output rule. This ensures that all
lexical entries that categorize for the input rule also subcategorize for the
output rule. The Head Feature Convention copies the features MC and
FIN onto the head of the right-hand side, i.e., onto the verb. Rule
(126a) would playa role in accepting (127a), rule (126b) in accepting
(127b).
(127)
a. kennt der Mann das Buch
+FIN +NOM +DAT
knows the man the book
b. dem Kind ein Mann hilft.
+DAT +NOM +FIN
the child a man helps
The accepted clause (127a) could serve as an alternative question with
the meaning Does the man know the book? Clause (127b) could com-
bine with a subordinate conjunction such as weil (because) to form the
well-formed subclause (128).
(128) ...weil dem Kind ein Mann hilft.
because the child a man helps
... because a man is helping the child.
Notice, that all permutations of the NPs can occur in rules since all the
LP rules proposed so far do not constrain the linear order of NPs.
Rules (129a) and (129b) are ill-formed and therefore cannot be
derived.
(129)
a. (5, V3 -+ V N
2
)
-MC +NOM
-FIN
b. (5, V3 -+ V N
2
)
+MC +NOM
-FIN
Rule (129a) violates LP (125b); (129b) violates FCR.
This does not preclude the possibility of rules for nonfinite clauses.
Rule (130) can be generated.
(130) (5, V3 -+ V N
2
)
-MC +NOM
-FIN
Although example (131) which is generated by (130), is ill-formed as a
sentence, rule (130) is needed for the derivation of acceptable sentences
with auxiliaries.
(131) *Peter kommen.
Two subsequent analyses of selected aspects of German word order
make also use of the structure-flattening approach. However, both Ner-
bonne (1982a)29 and Russell (1983)30 restrict structure flattening to verb-
initial and verb-second clauses. Thus, they lose an important generaliza-
tion, i.e., that the type of constituent scrambling that motivates flat
clause-structures is a property of all German clauses and is in no way
dependent on the position of the finite verb. The subordinate clause of
the following example does not have a contiguous verb phrase:
29Nerbonne's treatment of topicalization will be discussed in Chapter 6, Section 3.5.
30Russell presents an interesting account of separable prefixes, which will be investigated in
Chapter 4.
133
(132) Er wuBte, daB die Pille dem Patienten ein Doktor gegeben
he knew that the pill the patient a doctor given
hatte.
had
He knew that a doctor had given the patient the pill.
2.4 Notes on the Semantics of VP and Clause Rules
Versions of GPSG differ in the way semantic rules (translations) are as-
sociated with syntactic rules. The two predominant approaches are
described in Chapter 1, Section 5. The version of the framework that
assigns semantic components to all ID rules in the derivational history of
a PS rule, is more redundant than the more recent approach by Klein
and Sag (1982), but is is less problematic to formalize. In this version
my basic rule (98d) is supplemented by a semantic rule:
(133) (8, V
2
~ V, V, N
2
, N
2
, V('N
2
')('N
2
.)
+ACC +DAT +ACC +DAT
The semantics is identical with Gazdar's (1982a) proposal for English
and based on the proposal of Montague (1974b). The function denoted
by the ditransitive verb is applied to the intensions of the direct and in-
direct objects, in this order. Metarule (99) also needs to be extended to
operate on the semantic as well as on the syntactic rule:
(134) V
2
~ X, I/t =} V3 ~ N
2
X, I/tCN
2
.)
+NOM +NOM
The variable I/t stands for the semantic translation associated with the
input rule. The semantic translation of the output rule will be a func-
tional application of the denotation of the verb phrase to the intension of
the subject NP.
31
It was pointed out in Chapter 1, Section 5 that some of the problems
for the Klein and Sag approach to semantics are given rise by the fact
that in some cases the semantic component has to cooperate with the LP
component to induce the appropriate functor-argument structure. The
discussed example was the semantics of ditransitive verbs. For English,
the semantic component has to be integrated with the linear precedence
component because all standard GPSG grammars do not mark con-
31The functor-argument structure that Gazdar (1981) and other GPSG grammars associate
with the combination of subject and verb phrase follows Montague (1974a) instead of the
PTQ grammar in (1974b).
stituents as to their grammatical function. Thus the semantics cannot
"know" merely on the grounds of syntactic information encoded in the
ID rule, which of the two objects has to combine with the verb first.
If the semantic component were applied to German ditransitive VP
rules or on flat clause rules, a similar problem would arise. The infor-
mation about the semantic types associated with the syntactic categories
will not suffice to determine the order in which the verb picks up its ar-
guments. Because of the permutability of the noun-phrases in German,
an interaction with the linear precedence component cannot be of any
help. However, the case marking of the noun-phrases contains all the
information that is needed to construct the semantic structure. The dif-
ferent strategies for English and German rule extension therefore just ex-
press a well-known difference between the two languages: English marks
the function-argument structure by linear order, German by inflection.
There are two potential objections. One concerns verbs with two ac-
cusative NP complements such as lehren. Obviously, case marking can-
not predict the appropriate semantic structure for VPs or clauses with
two accusative NPs. This question will be discussed in Chapter 6, Sec-
tion 3.1.1. It will be argued that a feature that marks the second accusa-
tive complement is independently needed to account for certain syntactic
restrictions.
The second objection is based on sentences in which ambiguous case
endings create a syntactic ambiguity. Consider the following example
containing a regular transitive verb:
(135) Dann hat das Nilpferd das Krokodil
then has the hippo the crocodile
Then the hippo bit the crocodile.
gebissen.
bitten
Both noun phrases could be analyzed as carrying either accusative or
nominative case. It is usually the order of the NPs that determines the
appropriate interpretation. This could mean that an interaction between
semantics and linear precedence is needed in addition to having the
semantic component be sensitive to case marking. I will argue in Chap-
ter 5 that it is not the syntax that lets speakers use the unmarked order
of noun-phrases to disambiguate. It will be shown that the use of the
unmarked order in these cases only represents a strong preference, which
in turn is an instantiation of a universal communicative strategy to avoid
ambiguities.
It can be concluded then, that the proposed rules are compatible with
both GPSG approaches to semantics. An extension of the semantic com-
ponent is required that makes it sensitive to case marking. The rules
introduced so far do not necessitate the addition of grammatical
functions to the set of syntactic primitives.
3 Auxiliaries and Modals
3.1 Auxiliary VP and Clause Rules
One of the most disseminal publications in GPSG is Gazdar, Pullum, and
Sag (1982a)(GPS). The phrase structure grammar of the English
auxiliary system proposed in this paper is based on a main- verb analysis
for auxiliaries. Phrase structure solutions to the linear sequence of
auxiliaries and modals, to VP fronting, negation, imperatives, VP dele-
tion, subject-auxiliary inversion, and other well-known' regularities a.r e
presented. The approach to the German auxiliary system that Iprefer to
assume at this point is modeled largely after GPS (1982a).
In this analysis, modals, auxiliaries, and the infinitival particle to <lore
distinguished from other verbs by a feature +AUX. The rules that intro-
duce those verbs are not very different from other basic VP rules. They
can be described by the following rule schema:
(136) (n , V
2
~ V V
2
)
0' 13
+AUX
Once again, the features symbolized by 0' and +AUX are copied onto
V by force of the Head-Feature-Convention. The nine rules that are col-
lapsed in (136) differ from one another by the values they assign to 0'
and 13. Some combinations, such as double modal constructions, are ex-
cluded not out by these rules, but rather by holes in the inflectional
paradigms. In this case, modals do not have nonfinite forms and there-
fore cannot be heads of a nonfinite VP. However, modals do combine
with bare infinitival verb phrases.
I will not include here the full list of auxiliary VP rules, nor will I
repeat the argumentation in favor of their analysis. The effects of the
proposed auxiliary VP rules can be described best by viewing a phrase
structure tree for a complex VP. A simplified version of their tree for
(137) is given in (138).32
(137) Kim must have been being bothered by Sandy.
32The notation has been adapted to the convention followed in the present anlysis.
66
(138)
V3
/~
N
2
V
2
I /\
Kim V V
2
I /\
must V V
2
I /~
have V '12
I .>.
been V V
2
I~~
being bothered by Sandy
I assume an equally nested hierarchical structure for German clauses
containing auxiliaries and modals. It will be shown that the necessary
basic auxiliary VP rules, just like the other basic VP rules discussed in
Section 2.1, do not have to differ much from their English counterparts.
A simplified rule of this kind is:
(139) (1, V
2
~ V, V
2
)
+PERF +PSP
+AUX -PERF
(haben, sein)
This rule fits the rule schema in (136). As the rule stands, it states
that a perfective (+PERF) auxiliary VP can immediately dominate a
form of the auxiliary haben or sein, plus a non perfective VP headed by a
past participle (+PSP).33 However, it is not only verb phrases that con-
tain auxiliaries, but also clauses whose flat structure is due to the effects
of metarule (99) on page S1. Remember that these flattened clauses do
not have a matrix VP. Another metarule is needed that generates rules
for combining auxiliaries and clauses:
(140) V
2
---+ V, V
2
=} V3 ~ V, V3
+AUX
33The choice between haben and a second perfective aspectual auxiliary, 8ein, will be
discussed in Section 3.3.
Auxiliaries and Modals
137
The application of (140) to (139) yields:
(141) (1, V3 __,. V, V
2
)
+PERF +PSP
+AUX
(haben, sein)
The linearization of ID rules (139) and (141) is determined by LP
rules (125a) and (125b) on page 60. Depending on the value of the fea-
tures FIN and MC, the following PS rules derive from (141).
(142)
a. V3
~V V
2
+PERF +PSP
+AUX
+FIN
+MC
b. V3
__,.V
2
V
+PERF +PSP
+AUX
+FIN
-MC
c. V3
__,.V
2
V
+PERF +PSP
+AUX
-FIN
-MC
(By virtue of the Head Feature Convention, all features of the left-hand-
side symbol are copied onto the head unless indicated otherwise.)
A nonfinite auxiliary cannot precede the VP because the feature
cooccurrence restriction (124b) requires +MC constituents to be +FIN.
The ID rules (143), (144), and (145) are also simplified instantiations of
schema (136).
(143) (2, V
2
__,. V, V
2
)
+AUX +BSE
(muss en, konnen, dur fen, ... )
(144) (3, V
2
~ V, V
2
)
+AUX +PAS
(werden)
(145) (4, V
2
~ Y, y
2
)
+AUX +BSE
+FIN
(werden)
ID rule (143) combines modals with bare infinitival (+BSE) verb
phrases. ID rule (144) introduces the passive auxiliary werden by com-
bining it with passive (+PAS) verb phrases. The last ID rule, (145), lets
a finite verb phrase immediately dominate a form of the future auxiliary
werden together with a bare infinitival verb phrase. Metrarule (140) will
again provide the auxiliary VP rules with the corresponding clause rules.
These clause rules will closely resemble ID rule (141). They can be
represented by the following schema:
(146) (n, V3 ~ V, V3)
0' 13
+AUX
Anyone of these rules can introduce the finite verb of the sentence.
The finite verb needs to agree with the subject in person and number. In
Section 2.1 on pages 52 ff., it was described how the clause node, directly
dominating the subject, will get the agreement feature values of the sub-
ject. The Head Feature Convention in a traditional feature system
would not automatically percolate the agreement features up the tree, be-
cause it is always the auxiliary and not the clause that is the head of the
rules of type (146). However, the features will be carried up by perform-
ing the agreement on every level in the tree. To apply the agreement
unification at every level is not an added trick to substitute for the Head
Feature Convention, it is merely the result of not restricting agreement
to the subset of the auxiliary rules that introduces finite auxiliaries. The
agreement is simply called for by the Control Agreement Principle.
3.2 Some Consequences and Results of the Auxiliary Rules
The auxiliary rules introduced so far do not constitute a complete
analysis of the German auxiliary system. It would go beyond the scope
of this book to develop an in-depth analysis such as the one given by
Gazdar, Pullum, and Sag for English (1982a).34 However, the following
examples show that the rules perform quite a bit of work. The first ex-
ample (147) requires all four auxiliary rules.
(147) Wird Peter gesehen werden konnen?
will Peter seen be can
Will one be able to see Peter?
34Such well-known phenomena as double infinitives, subjunctives, negation, or the In-
finitival particle zu have to be disregarded.
Auxiliaries and Modals 69
V3
(148) v"
v =<>;
I v" I
wird / ~ konnen
V3 I
/\ werden
N
2
V
I geSihen
Peter
Compare the constituent structure indicated by the tree (148) with
the structure (138) of the English example (137). The auxiliary group
exhibits a left-branching structure, in contrast to the typical right-
branching structure in English verb phrases. Only the combination of
the clause-initial finite verb reminds of the English branching direction.
At the most one verb can occupy first position, because all auxiliaries
and modals combine exclusively with nonfinite constituents and it is im-
possible for a nonfinite constituent to have an initial verb.
Note that the clauses that can be generated in accordance with the
grammar comply with the pattern (44) in Chapter 1, Section 3 that sum-
marizes the ordering regularities within the auxiliary string, here
repeated as (149).
(149) MV (PA) MA* (AA) MA* (FA)
MV - main verb
PA - passive auxiliary
MA - modal auxiliary
AA - aspectual aux.
FA - future tense aux.
(geben)
(werden)
(konnen)
(haben)
(werden)
The passive auxiliary has to be closest to the main verb because it
combines only with constituents headed by a passive verb. Only main
verbs can have passive forms. This rules out ill-formed sentences such as
(150) or (151).
(150) *Peter wird sehen gekonnt.
Peter is see could.
Peter could be seen.
(151) *Peter wird gesehen gehabt.
Peter is seen had
Peter has been seen.
70
The future auxiliary werden, on the other hand, will always be far-
thest apart from the main verb, since it has to be finite and no auxiliary
or modal combines with a finite constituent. Therefore, (152) cannot be
generated.
(152) *Kann Peter kommen werden.
can Peter come will
Is it possible that Peter will come.
This has the consequence that the future auxiliary cannot appear In In-
finitival complements, which turns out to be a valid prediction.
(153) *Peter glaubt kommen zu werden.
Peter believes come to will
Peter believes he will come.
The inacceptability of (153) coincides with the ungrammaticality of the
corresponding English sentence (154):
(154) *Peter believes to will come.
However, the difference between English and German becomes apparent
when the acceptability of modals in the same position is considered.
(155) Peter glaubt kommen zu konnen.
Peter believes come to can
Peter believes that he can come.
(156) *Peter believes to can come.
Modals and perfective auxiliaries are not restricted as to their posi-
tion in the auxiliary string. They just need to combine with the right
type of VP. In most cases the appropriate type of the VP will be deter-
mined by a feature contributed by the head of the phrase. This is true
for features such as BSE or PSP. The feature PERF is different in
this respect. It prevents a verb phrase already marked perfective from
combining with another perfective auxiliary. This restriction makes it
impossible to generate (157a), (157b), and (157c).
(157)
a. *Hat Peter rufen haben.
has Peter call have
b. *Hat Peter kommen sein.
has Peter come have
Auxiliaries and Modals 71
c. *Hat Peter gerufen haben gekonnt.
has Peter called have could
As (157b) shows, it does not have to be a duplication of the same
perfective auxiliary that makes the sentence ungrammatical. Ru jen
combines only with haben, kommen only with sein. The second perfec-
tive auxiliary does not even need to be next to the first one to rule out
the sentence. Obviously, a head feature would not help. The clause
Peter gerufen haben gekonnt could not be marked +PERF if haben in-
troduced the feature as a head feature, since haben is not the head of the
clause. The feature PERF needs to be a foot feature that would be
carried up all the way to the root of the structure. Only then could the
restriction in rule (139) that allows haben and sein to combine only with
-PERF verb phrases prevent sentences (157a) and (157b) from being
generated.
It is not too surprising that a phrase structure approach could
provide an observationally adequate solution to the syntax of the Ger-
man auxiliary system. After all, the pattern that encodes the language
to be generated is a regular expression. It is noteworthy, however, that
the proposed grammar in its category inventory (including features) uses
only concepts that naturally correspond to observable phenomena.
Categories such as V, V
2
, or V
3
, or features like FIN or PSP are con-
ventionally assumed in linguistic description; and even the more special-
ized feature PERF, which marks the perfective aspect, corresponds to a
concept that can be found in any descriptive grammar of German.
3.3 Main Verb Analysis Versus Flat AUX Node
Gazdar, Pullum, and Sag (1982a) provide a nice summary of arguments
in opposition to the affix-hopping solution of Chomsky (19~)7;
1965) which is based on a flat structure for the auxiliary group. Many of
these arguments carryover to the German auxiliary system. Coupled
with the satisfactory results of the analysis presented in the previous
paragraph, they should constitute sufficient justification for my ap-
proach. I will not repeat here the evidence against the flat AUX node
analysis presented in the literature on main verb analyses, e.g., Pullum
and Wilson (1977) and Akmajian, Steele, and Wasow (1979).
Arguments against this analysis fall into several categories. Some
are just directed against the concomitant transformation of affix-hopping
(Chomsky's (1957) Auxiliary Transformation). Others criticize the as-
sumption of a flat AUX node, of an AUX outside the VP, or of an AUX
node altogether. The piece of evidence from German I want to present in
support of a hierarchical structure for the German auxiliary string is
directed mainly against the assumption of a flat auxiliary structure in a
phrase structure based framework, although the argument carries over to
the transformational paradigm. It does not matter for my considerations
whether the flat structure is dominated by an AUX node or integrated in
the VP. It will become clear later during the discussion of some truly
problematic data involving the fronting of strings that include auxiliaries
that the following brief argumentation is more than just "an exercise in
shadowboxing."
It is quite obvious that the syntax of the English auxiliary system
describes a finite language. Remember the AUX rule in Chomsky (1957),
here repeated in a slightly simplified version:
(158) AUX ~ (M) (have+en) (be+ing)
Even if this rule were extended to include other auxiliaries, such as do
and passive be, the resulting language would still be finite because none
of the elements, all of which are optional, could be repeated. The pat-
tern for the German auxiliary string can be written as a regular
expression (see (149) on page 69). In both languages all auxiliary strings
could be generated in a flat structure-in English by means of simple PS
rules, in German by employing the Kleene star notation that is used, for
instance, by Gazdar (1982b) for a similar purpose (i.e., to create flat
structures for coordinate conjunction).
Many of the arguments against the flat structure approach are based
on the complex relationships that exist among the auxiliaries present in
the auxiliary string. The presence of a certain auxiliary often determines
the inflection of an adjacent auxiliary. Sometimes, as in the case of the
perfective auxiliaries in German, it also rules out the concurrent presence
of certain other auxiliaries. The transformation of affix-hopping encodes
a small subset of these interdependencies. A flat auxiliary structure in a
phrase structure based framework has to express the interdependencies in
an intricate system of selection features that must be associated with the
symbols on the right-hand side of the flat AUX rule. In any-main verb
approach, already existing mechanisms for expressing subcategorization
and selectional restriction can be used to encode the complex relation-
ships. A flat structure approach needs to exploit the notion of adjacency,
as done by affix-hopping. This reliance on linear order alone would
render the option useless for the ID/LP format of GPSG, since ID rules
do not contain any information about linear order.
In a main-verb approach, only the dominance relation determines
which auxiliaries can select which. This is extremely helpful in the
analysis of German, since one member of the auxiliary string-i.e., the
finite verb-does not have to be adjacent to the other auxiliaries. The
following brief discussion of one of the dependencies between German
auxiliaries might serve as to illustrate the superiority of the main-verb
approach.
German, like some Romance languages, has two auxiliaries that
Auxiliaries and Modals
73
mark perfective aspect. Some form of the auxiliaries haben or sein com-
bines with the past participle of a verb to mark it as perfective. ID rule
(139) is the source of the PS rules that perform this combination. The
choice between the appropriate forms of haben and sein depends on the
verb that is modified by the auxiliary. It is inessential for the present
argumentation whether each verb needs a specific feature that indicates
with which of the two auxiliaries it combines or whether the choice is
determined by a configuration of other syntactic and semantic properti.es
of the verb.
35
The verb such en combines with haben, as in (159a) and
(159b).
(159)
a. Peter hat gesucht.
Peter has searched.
b .... , weil Peter gesucht hat .
.. .because Peter has searched.
If the verb is passivized, as in (160), the perfective auxiliary is a form of
setn.
(160)
a. Peter ist gesucht worden.
Peter has searched been
Peter has been searched for.
b .... , weil Peter gesucht worden ist.
because Peter searched been is
... because Peter has been searched for.
The main-verb analysis provides an easy solution. Whatever features
determine the choice of the appropriate auxiliary, they will be head
features of the verb. In (159) gesucht is the head of the constituent that
hat combines with, while in (160) it is worden. The solution will work
because forms of werden invariantly combine with forms of sein.
35
An
analysis based on a flat AUX node could simply stipulate that the feature
+PASSIVE on the AUX always forces the combination with sein. This
might be formulated as a feature cooccurrence restriction. However, it is
not the mere presence of a passivized verb or of the passive auxiliary that
triggered the choice of the perfective auxiliary in (160). Consider (161):
35There exist regularities that depend mainly on whether the verb is transitive or intran-
sitive and whether it is stative or not. However, regional variation and lexical exceptions
might require introduction of the special selection feature.
35This is even true for the main verb werden (become, turn into) which also might histori-
cally explain the choice in the case of the passive auxiliary,
(161) Peter hat gesucht werden sollen.
Peter has searched been should
Peter was supposed to be searched for.
This time the correct perfective auxiliary is a form of haben although
the main verb is passivized. Proponents of the main verb analysis do not
have to search for an explanation; it falls out from the fact that all con-
stituents with sollen as the head can combine only with haben. Thus,
the selection of the correct perfective auxiliary in German adds one more
argument to the long list of advantages of using a main-verb analysis for
auxiliaries.
3.4 Some Notes on the Semantics of the Auxiliary System
Auxiliaries and modals (at least in their epistemic reading) semantically
modify propositions. The semantics of the GPS analysis takes this into
account. GPS was based on the version of GPSG that associates seman-
tic components with all protorules. The auxiliary rule schema introduced
here as (136) is repeated in (162), this time with the semantic translation
schema.
(162) (n; V
2
__,. Y, V
2
; )..~[VCV2(~J)
0' 13
+AUX
Metarule (140) obviously needs to supply its output rules with a different
translation, since those rules combine an auxiliary with a clause. The
desired translation is easily stated:
(163) (n; V3 ~ V, V
2
; V' Cy2)
0' 13
+AUX
However, this could be a problem for the newer version of GPSG
that employs a special component of the metagrammar for generating of
semantic translations.
The algorithm that attaches the semantic component to ID rules
works on the basis of the syntactic information encoded in the ID rule.
It makes use of a small set of semantic operations to combine the trans-
lations of the constituents. In the majority of cases, functional
application is used. Other operations correspond to raising and equi
constructions.
In the present case, functional application could be used either for
rules of type (162) or (163), but not for both. In the former case,
auxiliaries and modals would be of type (V2, V
2
), in the latter of type
Auxiliaries and Modals ~'5
(V
3
, V
3
). Of course, it might be suggested that modals and auxiliaries
could be of different types. However, one would like to avoid a sys-
tematic ambiguity involving all auxiliaries and modals.
A better solution is based on Sag and Klein's (1982) treatment of
modals and auxiliaries as raisingf verbs. The semantic translation
schema in (162) corresponds closely to the semantics of raising construc-
tions. Raising verbs have been regarded as denoting functions from
propositions to propositions, in just the same way as this is done by
auxiliaries and modals. Under this assumption, the metagrammatical
component that assigns semantic translations to ID rules will combine
auxiliaries with clauses by means of functional application, but it will
make use of the semantic operation designed for raising constructions
when auxiliaries combine with verb phrases.
4 Verb-Second Clauses
The grammar assembled so far generates sentences of two out of three
clause types: verb-initial and verb-final clauses. In this section I want to
extend the grammar to accommodate verb-second clauses, the pattern
most German main clauses follow.
Let me repeat here and summarize the most important facts about
verb-second clauses, which were described in greater detail in Chapter 1,
Section 2.2 and Chapter 3, Section 4:
Assertion main-clauses and constituent questions main clauses
have the finite verb in second position.
There are also some subordinate clauses that exhibit the verb-
second pattern.
Drach's Law states that every major clausal constituent can
precede the finite verb in verb-second clauses.
In constituent questions, the question-marked constituent oc-
cupies first position.
In both constituent questions and assertion main clauses, there can
be an unbounded distance in the tree between the fronted constituent
and the gap it leaves behind.
(164) Von welchem Sanger hast du Peter gebeten
from which Singer have you Peter asked
zu versuchen, ein Autogramm zu bekommen?
to try an autograph to get
From which singer did you ask Peter to try to get an autograph?
(165) Von diesem Sanger hast du Peter gebeten
from this Singer have you Peter asked
zu versuchen, ein Autogramm zu bekommen.
to try an autograph to get
You asked Peter to try to get an autograph from this singer.
I will treat all first-position constituents of verb-second clauses as fillers
for long-distance dependency gaps. This approach is not new; similar
analyses can be found in Diderichsen (1962) for Danish and Welin
(1979) for Swedish. Since Chomsky (1977) proposed that topicalization
and wh-fronting be viewed as instances of the same long-distance move-
ment rule, known ever since as Wh-Movement, there has been an ongoing
debate about the similarities and differences between the two phenomena.
I do not want to engage in a detailed discussion of the properties of
topicalization and question formation in German. I will merely
demonstrate and motivate my use of GPSG's mechanism for long-
distance filler gap dependencies for the German verb-second clauses in
my fragment. In doing so I will, for historical reasons, retain the term
topicalization for the fronting in assertion clauses, although the term
sometimes conflicts with the standard terminology used for classifying
the discourse roles of constituents.
The rules that establish the filler gap dependency are of the following
form:
(166) V3 ~ 0', V3/0'
+AC +TOP +MC
+FIN
ID rule (166) is actually the topicalization rule. PS rules derived from
(166) will accept finite assertion main clauses (+AC) that are composed
of a topicalized constituent of category 0' and a main clause with a gap
of category 0'.37 The slash notation of Gazdar (1982a) is equivalent to
the newer slash-feature notation initiated by Bear (1981). S/O' is a com-
plex category symbol for sentences with a gap of category 0' somewhere
inside. An LP rule stating that the topicalized constituent comes first in
the resulting PS rules is necessary.
(167) +TOP < X
Rule (168) is a simplified version of the trace introduction metarule
which is essential for deriving PS rules that generate gaps.
37The discussion of topicalizations in subordinate clauses needs to be postponed until
Chapter 5, Section 3.5.
Verb-Second Clauses
77
(168) A ~ X,B =} AlB __,.X, t
A and B are variables over individual categories, X and Y variables
over strings of symbols. The slash notation again indicates a category
with a gap, while AlB stands for a category that is exactly like A except
that a constituent of category B is missing somewhere inside A. One can
now use the rule derivation mechanism in Gazdar (1982a) or, as in more
recent versions of GPSG, the regular feature instantiation mechanism, to
create the necessary rules for linking the filler with the gap.38 All these
rules, which accept the nodes in the tree that are located on the path
from the filler to the trace, are of the following form:
39
(169) c!D ~ X, E/D
Despite the vast literature on the pragmatic background of
topicalization, there is little agreement about the exact conditions that
trigger topicalization in English or that determine the fronted constituent
in German. I will return to this problem later in Chapter 6, Section 3.5.
For the time being I let me simply assume that one or several feature
cooccurrence restrictions ensure that the feature +TOP will always coin-
cide with a set of pragmatically determined syntactic features that
qualify the constituent for being fronted.
4o
Other feature cooccurrence restrictions will make sure that only con-
stituents of the appropriate syntactic categories are fronted. The con-
straints expressed by these feature coocurrence restrictions often cor-
respond to similar conditions in English. Other restrictions correspond
to bounding properties or island constraints.
For descriptions and discussions of the GPSG approach to long-
distance dependencies, and for refinements of the slash introduction and
propagation mechanism, see Gazdar (1982b), Gazdar and Pullum,
(1982a), Flickinger (1983), and Pollard, and Sag (1983).
The only statements I will make at this point about the syntact-ic
properties of topicalization in German concern some drastic differences
between the German phenomenon and its English counterpart that have
38The idea of using special categories to carry holes up the tree has been first discussed by
Baker (1978)(pp.108-115). Baker rejects the proposal in favor of a transformational
analysis, The strategies GPSG uses for subcategorization and feature percolation allow
for an implementation of the phrase structure approach to unbounded dependencies t.hat
is not viable to Baker's original criticism,
39 Actually, the right-hand side can contain more than one slash category, but this is of no
relevance for the present discussion.
40For the interaction of syntactic features and pragmatics, see Chapter 5.
78
to be considered in setting up rules for the fragment. In contrast to
English, unbounded dependencies in German have traditionally been con-
sidered clause-bounded. This is supposed to mean that there is no
extraction from finite S constituents. Counterexamples are discussed in
BoasY They usually involve the extraction of adverbial phrases.
(170) In welchem Jahr sagte Peter eben, daB er sie zum
in which year said Peter just that he her to
letzten Mal gesehens hat?
last time seen has
Which year did Peter J'ust now mention for having seen her
for the last time?
For some speakers of German, but not for me, even the following depen-
dency is acceptable.
(171) Das ist genau die Losung, die ich glaube, daB
that is exactly the solution which I believe that
eines Tages gefunden wird.
one day found will.
That is exactly the solution that I believe will be found one day.
The judgments vary immensely, sometimes in correlation with age
and regional origin. It is difficult to say to what extent positive judg-
ments are influenced by regional dialect, some of which are less restricted
in their boundary constraints than standard German. As I am not aware
of any conclusive results that might settle the question, I will make no
provisions for clause boundedness in the grammar for the fragment.
42
There is no preposition stranding in German.
(172) For this book Peter looked several times.
(173) This book Peter looked for several times.
(174) Nach diesem Buch suchte Peter mehrere Male.
after this book looked Peter several times
For this book Peter looked several times.
41Talk at The University of Texas in July 1979.
42The implementation of clause-bonded ness would be simple. A simple feature cooccur-
rence restriction that rules out a slash feature for tensed subclauses would suffice.
Verb-Second Clauses :r9
(175) *Dieses Buch suchte Peter nach mehrere Male.
Contrary to English, German allows for the extraction of nonfinite main
verbs.
(176) *Read Peter will the book.
(177) Lesen wird Peter das Buch.
In order to account for the last two differences I will modify the
trace introduction metarule. The modification in (178) is suggested for
the sole purpose of obtaining an adequate grammar for the fragment.
(178) V3 __,. X,B =} V3/B ~ X,t
-AUX -AUX
Metarule (178) will guarantee that only main verbs or their siblings can
be topicalized. The data introduced and discussed in Chapter 6, Section
3.5 on topicalization in German will show that the rule, as it stands now,
is too restricted. For the time being I will postpone any further discus-
sion of the first position in German main clauses until that section and
conclude the present subsection with a demonstration of the new rules.
Metarule (178) will derive three new ID rules from ID rule (179):
(179) V3 ~ V, NP, NP
+ACC +NOM
The new rules are given below under (180a)-(180c):
(180)
a. V3/V __,. NP, NP, t
+ACC +NOM
b. V3/NP ~ V, NP, t
+ACC +NOM
c. V3/NP -+ V, NP, t
+NOM +ACC
The feature instantiation mechanisms of GPSG provide us also with a
new set of rules of the form (182), derived from rules that match the
schema (146), repeated here in a slightly simplified notation as (181)Y
43In older versions of GPSG, (182) would have been derived by the slash rule derivation
mechanism.
(181) V3 __,. V, V3
+AUX
(182) riO' ~ v, V3/0'
+AUX
The interaction of the derived rules and the topicalization rule (166)
yields, among others, the following verb-second sentences:
(183)
a. Lesen wird Peter das Buch.
read will Peter the book
Peter will read the book.
b. Das Buch wird Peter lesen.
the book will Peter read
c. Peter wird das Buch lesen.
Peter will the book read
Chapter 4
Separable Prefixes
1 Introduction
One of the most interesting topics in the grammar of German verbs is
the problem of separable prefixes. The variety of terms for this
phenomenon indicates the lack of consensus as to a theoretical explana-
tion. Some of the more frequently used names for the prefixes are verb
particles (Verbpartikeln) and verb adjuncts (Verbzusatze). The process
of combining such prefixes with the verb stem is referred to as
prefixation, compounding, contraction, or incorporation. The multitude
of technical terms not only reflects the diversity of views about complex
verbs in German, but also stems from the frequent assumption that
several processes contribute to the formation of complex verbs.
The phenomenon demands attention within the domain of word or-
der research because the prefix can be separated from the verb stem.
Moreover, the cases in which the prefix separates from the stem depend
on the type of the clause with respect to the position of the finite verb.
It is precisely in those cases in which a finite main verb occupies first or
second position in the clause, that the prefix takes the ultimate position
in the clause core.
(184)
a. Peter wird Paul anrufen.
Peter will Paul up-call
Peter will call Paul.
b .... , weil Peter Paul anruft.
because Peter Paul up-calls
... because Peter calls Paul.
c. Peter ruft Paul an.
Peter calls Paul up
Peter calls Paul.
81
82 Separable Prefixes
d. Ruft Peter Paul an?
calls Peter Paul up
Does Peter call Paul?
An obvious point of debate is whether a separable prefix and a verb com-
bine to form a new word in the syntactic structure. Since it is the pur-
pose many of my examples to illustrate this debate and since words can
usually be recognized as strings of letters between delimiter characters,
including punctuation marks and blank spaces, a word of caution is in
order.
It is very important not to get confused here by orthographic con-
ventions. In my examples I will try to adhere to the current rules of
written Standard German as much as possible. In some cases, however I
will need to neglect those rules to demonstrate the intended word
boundaries. Rule violations will be restricted to word boundaries and
capitalization. In general, one cannot rely on the standard orthography
as an indicator for category hood or constituent structure. This is even
more true in the area of German complex verbs. The relevant or-
thographic rules and conventions are changing rapidly and are far from
consistent. This is one of the few areas in which most, and probably all,
native German speakers-the arbiters of orthographic standards
included-experience great difficulty in relying on their linguistic intui-
tion. It has been pointed out that native speakers usually have a very
precise feeling for word boundaries, despite the lack of consensus among
linguists regarding the definition of the concept word. Well, this does
not apply to German complex verbs. Therefore, the only conclusion we
can draw from orthographic conventions in this case is that an adequate
analysis 0. rable prefixes should somehow account for the difficulties
of speakers L pecify word boundaries for complex verbs.
In presenting my analysis I will follow the pattern established in the
previous chapter. Thus, I will first introduce and demonstrate the rules
that enable my grammar to generate sentences like those under (184).
Then I will discuss and finally reject some alternative propos <also
Throughout the chapter I will use the term prefix for all elements that
are perceived as part of the verb if they directly precede the verb stem,
independently of their origin.
44
The reason for this convention is that
my syntax treats all separable verb prefixes alike. The syntax can
neglect the significant dissimilarities among different classes of prefix-
verb combinations because I assume a lexical analysis for all such com-
binatory processes.
44Inseparable (or morphological) prefixes are not considered here at all, since they are of no
relevance for the problem of word order,
Introduction 83
2 The Rules
Every complex verb in the lexicon carries a feature that indicates the
prefix. For the present grammar a single binary feature suffices that en-
codes only the terminal string associated with the prefix. In a larger
grammar with multi-valued features, the string could be the value of a
feature PREFIX. The general framework of GPSG allows for complex
feature values. Hence, one could extend the value of the PREFIX feature
to encode other properties of the prefix in addition to the terminal strin g.
There are thus at least two entries in the lexicon for rufen; one has the
feature +an and translates into caU(up), the other does not have this fea-
ture and translates into call, yell.
45
In this case it happens that the two
entries carry the same sub categorization feature, both subcategorize for
rule (98b), which is repeated below under (185).
(185) (6, V
2
~ V, N
2
)
+ACC
kennen, suchen, nehmen
(know) (search)(take)
Metarule (186) creates rules that introduce the prefix as a sibling to
the verb and its complements:
(186) V
2
~ X =} V
2
~ X, SEPREF
O'E{ab, an, ... , zu}
+0' +0'
A feature default instantiation principle'i'' needs to be stated that
sets the default value for the PREFIX to be NIL, empty, or minus. The
current formalization, which is restricted to binary features, would make
the statement of the principle unnecessarily messy, since it would have to
be stated individually for every single prefix feature, i.e., for every prefix.
The Head Feature Convention copies the prefix feature onto the
verb. The new constituent SEPREF now carries the feature that in-
dicates the individual prefix. One could restrict the metarule to apply to
main-verb phrase rules only, but no damage will be done if this is
45Both verb entries are associated with the same terminal string: call. (In a full-fledged
grammar with multivalued features, the terminal string is the value of a feature LEX.)
46Gazdar and GazdarPullum (1982a) use the term feature coefficient default (FeD) for
this kind of rule.
84
Separable Prefixes
omitted, since no other verbs will carry a prefix feature.
47
One of the
output rules of (186) will be (187), which is derived from (186):
(187) V
2
__,. V, N
2
, SEPREF
+an +ACC +an
The following ID rule expands the constituent SEPREF:
(188) SEPREF __,. 0'
+0'
A technique has been used here that has proved sucessful in dealing with
the prepositions of prepositional objects. The strategy of introducing lex-
ical items into the syntax by a feature whose value is the lexical string
itself was originated by Gazdar and Sag (1982).
One more rule is needed. LP rule (189) will ensure that separable
prefixes always follow all complements and adjuncts:
(189) X
2
< SEPREF
The existing rules will be sufficient to allow generation of the sentences
(184). Metarule (99) on 51 will derive clause rule (190) from (187):
(190) V3 __,. V
2
, N
2
, N
2
, SEPREF
+an +ACC +NOM +an
The LP rules permit the following (simplified) PS rule:
(191) V3 ~ V N
2
N
2
SEPREF
+an +NOtt +ACC +an
+MC
+FIN
This rule is the basis for generating (184d), repeated here as (192):
(192) Ruft Peter Paul an?
47If the principle of Finite Closure is replaced by some other constraint(s), the left-hand-side
of the input rule had to be marked -a to avoid productive rules with multiple SEPREF
constituents. This would still not prevent unproductive rules with several SEPREF
constituents from being generated. Each of the SEPREFs would have a different prefix
feature. Verbs would not fit these rules because no verb entry has more than one prefix
feature, The unproductive rules could easily be filtered out by a feature cooccurrence
restriction. Obviously, a multivalued feature PREFIX would simplify matters here.
The Rules 85
Sentence (184c) is related to (192) by topicalization. The relevant clause
rule for (184b), repeated here as (193), is (194):
(193) weil Peter Paul anruft.
(194) y3 ~ N
2
N
2
SEPREF V
+an +NOM +ACC +an
-MC
+FIN
Now, there seems to be something missing here; after all, anruft is
written as one word and an is called a prefix, The derivation that fol-
lows from the present proposal, however, treats the prefix and the verb
as separate syntactic constituents. Russell (1983) suggests a morphologi-
cal rule that combines prefixes with -MC stems (-INV in Russell's
. ) 48
notation .
There are two reasons that are most often cited as evidence for con-
sidering prefix-verb combinations words. One is the accent on the prefix.
Any compound rule, morphological or syntactic, would account for the
accent, because German compounds are stressed on the first component,
Could an analysis that does not make the complex verb a single word
account for the accent on the prefix? In a sentence that has the nonfinite
main verb preceded by a complement or adjunct, the primary stress
would fall on the element before the verb, provided that the intonation
pattern is unmarked.
(195) Er hat das Geld geliehen.
he has the money borrowed
He has borrowed the money.
Contrastive or emphatic stress can also be put on th~ verb:
(196) Er hat das Geld geliehen (nicht gestohlen) .
he has the money borrowed (not stolen)
He has borrowed the money (not stolen it).
Contrastive stress could also be assigned to the stem of a complex verb:
(197) Er hat es abgenommen (nicht abgerissen) .
he has it down-taken (not down-torn)
He has taken it down (not torn it off).
48Russell treats prefixes as independent lexical items. I will compare this assumption with
the feature solution later.
86
Separable Prefixes
The reason that one cannot stress the stems of all complex verbs has to
do with their semantics. The stem often does not carry an autonomous
meaning:
(198) *Peter wollte teilnehmen.
This is parallel to the stress rules for idioms,
(199) Er wollte ihnen den Garaus machen.
he wanted them to kill
He wanted to kill them.
(200) *Er wollte ihnen den Garaus machen.
The second reason for calling complex verbs words originates with
derivational morphology. All German verbs can be nominalized. The
noun is always neuter and has the form of the infinitive: das Fahren (the
riding), das Kommen (the coming), das Rufen (the calling), All complex
verbs can be nominalized in this way: das Busfahren (the bus-riding),
das Runterkommen (the coming down), das Anrufen (the calling-up).
Some, but not all prefix verbs enter into other types of nominalizations:
die Busfahrt (the bus ride), der Anruf (the phone call). There is a noun
die Ankunft from ankommen, but no noun die Runterkunft from
runterkommen. If nominalizations of both kinds were restricted to apply
only to one-word verbs, the question would be settled. However, it is a
fact about German that also complement-verb, adjunct-verb, or adverb-
verb combinations can be nominalized: das Naseputzen (the nose
blowing), das Aufdensteinensitzen (the sitting on rocks), das
Schnellerfahren (the going faster). There are no complex verbs that
could be the source of those deverbal nouns:
(201) *Peter putzte nase.
Peter cleaned nose
(202) *Peter wird aufdensteinensitzen.
Peter will on-the-rocks-sit
(203) *Peter kann schnellerfahren,
Peter can faster-go
There is no reason to assume that the nominalizations involving prefixes
could not be formed by whatever process is responsible for deverbal
nouns that contain complements and adjuncts. I have not been able to
find any conclusive arguments against the hypothesis that prefix and
verb do not combine to form one word.
The Rules 87
But there is strong evidence against a morphological rule. I will dis-
cuss the relevant data when I compare my rules with the accounts of
separable prefixes proposed by Wunderlich (1983a) and Russell (1983).
At this point Iwill simply continue by suggesting not to treat the attach-
ment of separable prefixes on the subword level but rather to assume
some special contraction process that takes place only if the prefix ends
up as an adjacent sibling to the left of the verb.
49
This solution is in accordance with Wunderlich (1983a), who
proposes a rule of Inkorporierung (incorporation). I do not want to use
metarules or other existing metagrammatical tools to encode the
contraction process. Contraction depends mainly or solely on linear
adjacency. If metarules are made sensitive to linear precedence, an im-
portant generalization is lost. In addition, contraction processes can
apply across constituent boundaries. In the English examples (204) and
(205), the contracted words are not siblings.
(204) Who's next?
(205) There's your change.
A metarule could not contract who (204) or there (205) with is without
herwi . d f'latteni 50
some at erwise unrnotrvate structure attenmg.
One more argument against a metarule solution needs to be men-
tioned. For me it is still an open question whether any of these rules
have any effect on the syntactic structure or whether they merely
represent a very restricted phonological process. Therefore I do not as-
sume a syntactic reanalysis as Wunderlich (1983) does. The evidence to
be presented in the following subsections speaks clearly against a syntac-
tic contraction in the case of German separable prefixes.
I hope that the following discussion of alternative approaches rnignt
clear up any remaining doubts about the justification of my approach.
49 A similar phonological rule may be required for the contractions of prepositions and the
reciprocal pronoun einander:
Sie stritten mitelnander.
they fought with-each-other
They fought.
Die Bticher lagen tibereinander.
the books lay over-each-other
The books were lying on top o] each other.
50See Chapter 2, Section 7, for a hypothetical characterization of contraction metarules.
3 Alternative Proposals
To set the stage for a comparison of alternative approaches to the
problem of separable prefixes, I need to describe the phenomenon in a
little more detail. The origin of most German verb prefixes is trans-
parent. Nouns, adjectives, adverbs, and prepositions can become
separable prefixes (206), (207), (208), (209).
(206)
a. Peter wird radfahren.
Peter will bike-ride
Peter will ride a bike.
b. Peter fahrt oft rad.
51
Peter rides often bike
Peter bicycles often.
(207)
a. Peter wird das Haus sauberrnachen.
Peter will the house clean-make
Peter will clean the house.
b. Peter macht das Haus sauber.
Peter makes the house clean
Peter cleans the house.
(208)
a. Peter wird den Hut abnehmen.
Peter will the hat off-take
Peter will take off his hat.
b. Peter nimmt den Hut abo
Peter takes the hat off
Peter takes off his hat.
(209)
a. Peter wird vorgehen.
Peter will go-before
Peter will go earlier/in front of us.
b. Peter geht vor.
Peter goes before
Peter is going earlier/in front of us.
There are a few instances in which verbs become prefixes of other verbs.
5lIf a denominal prefix is separated from the stem, it is often capitalized like a regular
noun. (Peter [ahrt Rad.)
Alternative Proposals 89
(210)
a. Peter wird sitzenbleiben.
Peter wird sit - remain
Peter will remain seated.
b. Peter bleibt sitzen.
Peter remains sit
Peter remains seated.
Some separable prefixes such as dar, inne, and acht do not occur
anywhere else but in their role as prefix.
(211)
a. Peter wird ein Lied darbieten.
Peter will a song ? offer
Peter will per form a song.
b. Peter bietet ein Lied dar.
Peter offers a song ?
Peter per forms a song.
(212)
a. Der Gedanke des Friedens soll diesem Treffen
the spirit of peace shall this meeting
innewohnen.
in live
The spirit of peace shall pervades this meeting.
b. Der Gedanke des Friedens wohnt diesem Treffen inne.
the spirit of peace lives this meeting in
The spirit of peace pervades this meeting.
When the separable prefix occurs in conjunction with the verb, the
primary stress falls on the prefix. In this respect, the compound (or con-
tracted form) differs from otherwise potentially homonymous verbs with
inseparable prefixes.
(213)
a. Peter wird das Gepack ubersetzen.
Peter will the luggage over carry
Peter will carry the luggage over.
b. Peter setzt das Gepack uber.
Peter carries the luggage over
Peter carries the luggage over.
c. *Peter ubersetzt das Gepack.
Peter over carries the luggage
(214)
a. Peter wird das Buch ubersetzen.
Peter will the book translate
Peter will translate the book.
b. *Peter setzt das Buch uber.
Peter translates the book.
c. Peter ubersetzt das Buch.
Peter translates the book.
Sometimes the homonyms are almost identical in meaning.
(215)
a. Peter wird das Band durchschneiden.
Peter will the ribbon through cut
Peter will cut the ribbon,
b. Peter schneidet das Band durch.
Peter cuts the ribbon through
Peter cuts the ribbon.
(216)
a. Peter wird das Band durchschneiden.
Peter will the ribbon through cut
Peter will cut the ribbon,
b. Peter durchschneidet das Band.
Peter throughcuts the ribbon
Peter cuts the ribbon.
The verb in (216) is more restricted in use than the one in (215). The
sentence under (216) would be appropriate if the ribbon cutting were
part of an dedication ceremony; it would be less appropriate if the shor-
tening of a piece of Scotch tape is described.
Another peculiarity of separable prefixes in German is their behavior
with respect to the inflectional prefix ge- and the infinitival marker zu.
As in English, the past-participle marking indicates either passive voice
(217a) or perfective aspect (217b).
(217)
a. Peter hat das Buch gesucht.
Peter has the book searched for
Peter has sear-ched for- the book.
b. Das Buch wurde gesucht.
the book was searched for
The book was searched for'.
The past participles of inseparable prefix verbs are formed without the
Alternative Proposals
91
prefix ge-. Exceptions a.re very rare and always involve recent coinage of
verbs with nonstandard etymology.
(218) Peter fuhlte sich gebauchpinselt.
Peter felt self belly-brushed
Peter was flattered.
(219)
a. Der Doktor hat den Patienten untersucht.
the doctor has the patient examined
The doctor has examined the patient.
b. Der Patient wurde untersucht.
the patient was examined
The patient was examined.
Past-participle forms of separable prefix verbs are usually marked by ge-.
However, the past participle prefix in such verbs becomes an infix.
(220)
a. Peter hat aIle Regale abgesucht.
Peter has all shelves off searched
Peter has searched all the shelves.
b. Alle Regale werden abgesucht.
all shelves are off searched
All shelves are being searched.
The only exceptions are those few verbs that have an inseparable prefrx
in addition to the separable one, such as abbestellen:
(221) Peter hat das Abbonnement abbestellt.
Peter has the subscription cancelled
Peter has cancelled the subscription.
A similar infixation takes place if the infinitival zu combines with the
bare infinitive form of a separable prefix verb.
(222) Peter versuchte, die Regale abzusuchen.
Peter tried the shelves off to search
Peter tried to searcli the shelves.
This infixation is even more remarkable because the infinitival zu is not
considered a morphological element of other infinitival verbs.
92
Separable Prefixes
(223) Peter versuchte, das Buch zu suchen.
Peter tried the book to search
Peter tried to search for the book.
(224)
Der Doktor versuchte, den Patienten
the doctor tried the patient
The doctor tried to examine the patient.
zu untersuchen.
to examine
The two alternative analyses I want to consider here are Russell
(1983) and Wunderlich (1983a; 1983b). Russell's analysis and the one of-
fered here share the same grammatical framework while many of my
opinions about the phenomenon are at one with Wunderlich's views. In
fact, it is to Wunderlich that lowe many new insights into this matter.
Wunderlich cites the facts of infixation and the issue of separability
as two reasons for his doubts regarding the lexical status of the prefix
verb compounds. In many cases, the evidence is sufficient to rule out a
regular morphological prefixation process. The word accent on the prefix
provides the only indication for a joint word. The contrast between the
prefix stress on separable prefixes and the stem stress in regular prefix
verbs has been traditionally interpreted as the difference between com-
pound stress and regular word stress. This approach has been successful
regardless of whether German lexical stress assignment was considered
rule-driven Kiparsky (1966), lexical Wurzel (1970), or metrical Uszkoreit
(1979).
There remains then the possibility of viewing the relevant prefixation
phenomenon as a more or less standard compound formation process,
parallel to other compounding regularities in German. After all, German
seems to allow for all combinatorially possible binary combinations of
major lexical categories: AA, AN, AV, NA, ... , VV. Wunderlich
(1983a) investigates this analysis. He arrives at the conclusion that all
compounds of type XV, where X is A,N, or V, differ considerably from
the other kinds of combinations. For regular compound words there ex-
ists a variety of potential semantic relations among the meanings of the
components. One of Wunderlich's examples is the newly coined NN com-
pound Kanzlermontag (chancellor Monday). The word could denote the
Monday on which the chancellor is elected, inaugurated, or visited by the
members of the diplomatic corps. Or, to add Wunderlich's own examples
of potential interpretations, a Kanzlermontag could be the Monday on
which the chancellor can sleep longer, when he can accomplish his
greatest achievements, or it might be a Monday, on which all the school-
children sing a song dedicated to the chancellor. A new coinage of a
noun-verb compound kan zlereiurzen (chancellor-topple?) permits only
one interpretation: the action of toppling the chancellor.
Here is the way Wunderlich summarizes his findings about com-
pounds with verbs as second components:
52
The first component always binds a valency (argument
position) of the verb.
The first component is separable, i.e., there is only an
unstable combination.
A stable combination exists only in the case of
nominalization.
The observation concerning the special semantic properties of verb
compounding (or prefixing) is quite important. If the first component (or
separable prefix) always fills an argument position of the verb, we might
not need any rule for the combination at all. Indeed, Wunderlich sug-
gests a regular VP structure as the source of the peculiar compounds. He
accounts for the speakers' intuition about the association of argument
and verb he with a rule of reanalysis whose application he describes in
the following example:
Wunderlich names the rule incorporation (Inkorporierung). The rule
applies to other Head-complement compounds as well. An example
presented in the same paper illustrates the combination of prepositions
and nouns into complex prepositions (226):
(226) [pp [po in] [NP' .. Folge ... ]] -> [ppo [infolge] ... ]
Two possible objections to the fundamental observation come to
mind. One has to do with the tendency of more salient semantic rela-
tions to override other interpretations. The observation about verb com-
pound semantics might just follow from this tendency. Consider the
phrase Joe's team. Many relations can be expressed by the possessive
between the team and Joe: Joe can own or support the team, he can
work or play for the team, he can be coach, bodyguard, trainer, etc. The
phrase Joe's parents, however, has a much more predominant interprera-
tion: Joe is the son of his parents. Nouns such as father, daughter,
parents have often been analyzed as denoting two place relations. There
is a strong tendency for the possessive to be interpreted as denoting ex-
actly this relation if a possessive determiner is applied to such a rela-
tional noun. That does not mean that no other reading is conceivable.
If Joe is a pediatrician, the phrase Joe's parents can very well refer to
the parents of Joe's patients.
52Wunderlich, (1983b), p. 3, my translation
Wunderlich claims that nominalizations of verbs behave like verbs.
Kanelereturzer are people who are or were involved in toppling the
chancellor. No other interpretation is possible. This cannot be true for
all nominalizations. A Kanzlersturz does not have to be the toppling of
a chancellor, it could simply be a way of falling like a chancellor.
Kan.zleretur zer could very well be people who perform this kind of fall-
ing in downtown Bonn for American tourists. But, even if Wunderlich's
hypothesis does not apply to nominalizations of verbs, the generalization
still seems to hold for verbs. There are no examples that allow for a
variety of different interpretations depending on the assumed semantic
relation between the component meanings.
The second potential objection I would like to mention here concerns
the definition of valency required if Wunderlich's hypothesis is to be
maintained. Consider sentences (227) and (228):
(227) Peter wird autofahren.
Peter will car drive
Peter will drive.
(228) Peter wird busfahren.
Peter will bus go
Peter will go by bus.
If auto and bus fill argument positions, they must fill different argument
positions in the two sentences. 53 Wunderlich not only has to assume that
fahren in (228) is different from fahren in (227), but he also has to
classify mit dem Bus in (229) as an argument of the verb.
(229) Peter wird mit dem Bus fahren.
Peter will with the bus go
There is also the question of how the adverb ab or the preposition bei
can be seen as arguments of the verbs in (230) and (231).
(230) Peter wird abfahren.
Peter will off drive
Peter will leave/go away/drive off.
(231) Peter wird ein Buch beilegen.
Peter will a book by lay
Peter will add/include a book.
53Sentence (228) could also mean Peter will drive a bus if Peter has been hired as a bus
driver.
Alternative Proposals !l5
The problem is not the bare preposition bei, for Wunderlich's theory of
prepositions and prepositional adverbs can explain the lack of a NP. The
problematic aspect of Wunderlich's hypothesis is the necessity to consider
many constituents as arguments that so far have been classified as free
adjuncts. Ultimately, this could prove less of a problem for WunderIich
than for approaches that do not grant semantic argument role to adver-
bial phrases. Instead of speculating about this interesting problem here,
however, I will rather continue with my comparison of the alternative
approaches.
My analysis, which is based on Uszkoreit (1982), shares with
Wunderlich's the absence of a morphological rule for separable prefixa-
tion. However, I consider the stem and its separable prefix to be lexical
units. That is not to say that I disagree with Wunderlich's hypothesis
about argument incorporation. Nevertheless, this rule should be for-
mulated in terms of lexical rules, including word formation rules. As al-
ways in a confrontation between a syntactic and a lexical solution,
productivity and regularities have to be compared with idiosyncracies
and exceptions. In the case of separable prefixes, there is a wide range
spanning the whole distance between the two extremes.
Many verbs with separable prefixes have meanings that cannot be
derived compositionally from verb and prefix without creating additional
lexical entries.
(232)
a. Peter nimmt abo
Peter takes off/away
Peter loses weight.
b. Peter nimmt teil.
Peter takes part
Peter takes par-to
C. Peter teilt Paul ein Geheimnis mit.
Peter shares Paul a secret with
Peter tells Paul a secret.
Some prefixes, such as dar, inne, and acht do not occur outside complex
verbs:
(233)
a. Peter g1bt acht.
Peter gives attention
Peter is paying attention.
b. Peter halt inne.
Peter holds in
Peter pauses.
There are also verbs that are no longer found in the standard language
without separable prefixes:
(234)
a. Peter strengt sich an.
Peter exerts self?
Peter exerts himself.
b. Peter stattete Paul einen Besuch abo
Peter paid Paul a visit?
Peter paid Paul a visit.
Finally, there are many potential complex verbs that should be found
but do not actually exist:
(235)
a. Peter wird busfahren.
Peter will bus-go
b. Peter wird zugfahren.
Peter will train-go
Peter will go by train.
C. *Peter Wird schnellzugfahren.
Peter Will express-train-go
d. *Peter Wird kombifahren.
Peter will station-wagon-go
(236)
a. Peter wird skilaufen.
Peter will ski-run
Peter will ski.
b. Peter wird skifahren.
Peter will Ski-go
Peter will ski.
C. *Peter wird skigleiten.
Peter will Ski-glide
d. *Peter wird skirennen.
Peter will ski-run
The explanation for the variation seems simple. The acceptable com-
binations are based on frequency of use. Other factors, such as the
length of the prospective prefix, might also playa role.
(237)
a. Peter wird busfahren.
Alternative Proposals ~17
b. Peter wird mit dem Zubringerbus fahren.
Peter will with the shuttle-bus go
Peter will ride the shuttle bus.
c. *Peter Wird zubringerbusfahren.
Even frequent use of (237b) is unlikely to trigger the complex verb in
(237c).
At the other end of the spectrum are such very productive combina-
tions as compounds of directional adverbs and directional verbs-i.e.,
verbs whose lexical semantics implies or allows for the existence of some
local starting and end points for some activity or process.
(238) hinein
hinaus
hinuber
(into)
(out)
(over)
gehen (go)
sehen (see)
rufen (call)
Every combination of a prefix from the first column with a verb from
the second column yields an acceptable complex verb. The productivity
can be determined by either inventing new verbs or combining existing
verbs that not yet have a directional meaning with a directional prefix.
The verb in (239) is a new coinage. Its prefixation is instantly accept-
able.
(239) Peter wird hineinschwuppeln.
The verb grinsen (grin) is usually not a directional verb. Sentence (210)
makes it into a motion verb.
(240) Peter kann sich uberall hineingrinsen.
Peter can self everywhere into grin
Peter can grin his way into any situation.
An interpretation that immediately comes to mind is that Peter's grm-
ning is so irresistible that he can use it to gain entry anywhere.
Recent research in lexical frameworks (including Lexical Functional
Grammar) has demonstrated convincingly that productive processes can
be encoded in the lexical component. There is yet another indication for
separable prefixing to be a lexical phenomenon. The prefix often changes
the valency of the verb. Proponents of LFG have been quite successful in
gathering evidence for the hypothesis that all relation- changing processes
are lexical. If I do not attempt here to formulate lexical rules to account
for a good portion of the redundancy created by assuming lexical entries
for complex verbs, it is principally because of the lack of a fully
developed theory of the lexicon in GPSG.
It might have been the lack of a developed lexical component that
influenced Russell (1983) in suggesting his analysis of German separable
prefix verbs. Russell assumes a syntactic category PRF of separable
prefixes. In contrast to the syntactic category SEPREF in the analysis,
members of PRF are not derived in the metagrammar from a feature of
complex verbs; instead they are listed separately in the lexicon. There
are two modes of combination. Rule (241) is applied to the verb stem on
the sub-word level to yield a complex verb.
54
G!41) [V[LEX.-INVJ PRF V[-1]]
Semantically, the meaning of the prefix is applied functionally to the
meaning of the verb stem. In verb-initial and verb-second clauses, the
prefix is introduced by the metarule that creates the flat clause structure.
(242) ( ; v: -> W ; F ) =} ( ; V" [+1NV]
-+ W , N" [+NOM] , (PRF) ; F'(N" [+NOM] .))
Where:
(i)
(ii)
(iii)
(iv)
W is a variable over strings of category labels
and commas,
The optional PRF is a separated prefix,
F stands for the semantic portion of an 1DR,
F' is the result of substituting PRF'(V[-1]') for
V[O]' if V"[+1NV] -+ W, W, and F' = F otherwise.
Again, the prefix is the semantic functor. The choice of the appropriate
prefix-stem pairs is determined solely by the semantics of prefix and
verb.
In my critique of Russell's approach, let me just point out that his
choice to attribute the selection of acceptable prefix-stem pairs to the
semantics will disable him to explain the influence of frequency of use on
the creation of new pairs as well as the noncom positional frozen mean-
ings of examples (232), (233), and (234). Examples (235), (236), and
(237) cannot be accounted for at all. Let me demonstrate this lack of
explanatory power by extending (236) in (243).
(243)
Peter wird
skispringen.
skilaufen.
skifahren.
*skigleiten.
*skirennen.
*skigehen.
*skihiipfen.
54This is Russell's rule (17) in his own notation.
It is inconceivable that laufen, fahren, and springen could be in a
natural semantic class that does not contain gleiten, rennen, gehen, or
hup jen, Russell has no choice but to accept the existence of three dif-
ferent prefixes ski if he maintains his purely semantical approach.
It should also be noted tha.t Russell's nonlexical solution to the for-
mation of complex verbs does not spare him the additional complexity of
lexical rules as they are assumed in the current analysis, since one would
still have to link the different types of prefixes with the verbs, adjectives,
nouns, prepositions, and adverbs they are derived from.
A final remark on Russell's analysis concerns Wunderlich's
hypothesis that all separable prefixes are incorporated arguments of the
verb. Even if this hypothesis could not be maintained as a general rule,
it would still apply to a large subclass of complex verbs. Russell's func-
tion argument structure invariably makes the prefix the functor. The
analysis presented in this section is not committed to any single direction
of functional application. The mode of prefix-stem combination will
depend on the word formation rules, redundancy rules, and productive
lexical rules that need to be stated and, in many cases, remain to be dis-
covered.
The result of the comparison is easily summarized. Although each of
the three approaches has tempting features, only the solution introduced
here is an adequate analysis of the multitude of phenomena connected
with German separable prefix constructions. I believe I was able to
defend the lexical solution against both syntactic approaches. At this
point, let me briefly summarize the features of the three analyses and
then proceed to the last step in the comparison in which the topicaliza-
tion test will be used to determine the adequacy of the assumed con-
stituent structures.
From Wunderlich's standpoint, separable prefixes are nouns, adj ec-
tives, adverbs, or prepositions that function syntactically as complements
(or parts thereof) of the verb. A rule of reanalysis called incorporation
contracts prefix and verb. It is left open what triggers the contraction.
Russell views all prefixes as members of a separate lexical category. In
clauses in which the main verb is fronted to first or second position, the
prefix is introduced as a sibling to the verb and its complements; in all
other cases, the prefix is merged with the verb by means of a mor-
phological rule. The prefix denotation is always a function from verb
denotation to verb denotation. The acceptability of a complex verb is
determined solely on semantic grounds.
The lexical approach put forward in this book assumes a lexical
entry for each complex verb. The entries may be linked with the entries
of the components by different kinds of lexical rules. Consequently, the
choice of acceptable prefix-stem pairs is fully determined by the lexicon.
The prefix always becomes a sibling to the verb and its complements. If
the prefix ends up before the verb, a phonological contraction process
may apply.
4 Separable Prefixes and Topicalization
Maybe the most reliable test for constituency in German is topicalization.
In Chapter 3, Section 4, it was demonstrated that only one constituent
can precede the verb in assertion main clauses. Apparent coun-
terexamples discussed in Chapter 6, Section 3.5, adhere to this
generalization. Separated prefixes can often be topicalized.
(244) Runter kommt er immer. (runterkommen)
down comes he always
He always manages to get down.
(245) Auto fahrt er selten.
car goes he seldom
He seldom drives a car.
(246) Weiter geht es nicht.
farther goes it not
It doesn't go any farther.
(247) Heraus sprang ein junger Offizier.
out jumped a young officer
A young officer jumped out.
The corresponding topicalizations of nonseparated prefixes are much less
acceptable.
(248) ?Heraus war ein junger Offizier gesprungen.
out was a young officer jumped
A young officer had jumped out.
(249) ?Weiter kann es so nicht gehen.
farther can it so not go
It can't go on like that.
Topicalization of the complex verb as a whole is perfectly acceptable.
(250) Herausgesprungen war ein junger Offizier.
out-jumped was a young officer
(251) Weitergehen kann es so nicht.
on-go can it so not
However, there are enough cases of more acceptable topicalization of non-
separated prefixes to at least prevent us from drawing hasty conclusions.
Separable Prefixes and Topicalization 101
(252) Runter kann er immer kommen.
down can he always come
He can always come down.
(253) Auto kann er nur selten fahren.
car can he only seldom drive.
He can drive only seldom.
Other topicalizations of separable prefixes are bad in all cases.
(254) *Teil nimmt er immer.
Part takes he always
(255) *Teil kann er immer nehmen.
Part can he always take
(256) *Zu stellt er den Brief.
? deliver he the letter
(257) *Zu hatte er den Brief gestellt.
? had he the letter deliver
The contrast between topicalizations of types (250)-(253) and (254)-(257)
lies in their semantics. Runter, heraue, and weiter contribute much
more to the meaning than teil and zu; actually, it is highly unlikely that
the latter two carry any independent meaning at all. The fronting of
semantically non autonomous prefixes is even blocked in cases in which
the prefix establishes a contrast. The context for the following sentence
could be Peter's plan to enroll in a weight reduction workshop:
(258) *Teil kann er immer nehmen, mit dem Abnehmen
part can he always take with the weight-losing
sieht's schon schwieriger aus.
looks-it already more difficult
No doubt he can take part; losing weight is the hard part.
The only conceivable use of a sentence such as (258) is as a pun based on
its ungrammaticality. Yet the contrastive topicalization is acceptable in
the case of adverbial prefixes. This time the context is Paul's climbing a
tree.
(259) Rauf kann er immer klettern, mit dem Runterklettern
up can he always climb with the down climbing
sieht's schon schwieriger aus.
looks-it already more difficult
He can always climb up, but getting down is something else
aqain,
All three analyses can handle the extractions in (260). Wunderlich might
also be able to construct an account for the topicalization of non-
separated prefixes and complex verbs in (201). He could claim that
prefix extractions operate on the nonreanalyzed, structure whereas ex-
tractions of the complex verb operate on the structure after reanalysis
has taken place. There remains just one difficulty. Wunderlich's incor-
poration rule would derive busfahren as follows:
The deletion of accompanying material of the prospective prefix and the
reanalysis take place in one step. The topicalization in (261) does not
have a source.
(261) Bus ist er hochst selten gefahren.
bus is he extremely seldom gone
He went very seldom by bus.
Russell cannot account for the topicalization in (261), since prefix and
verb form a single word in his grammar. No evidence whatsoever can be
found for topicalizations out of words, not even out of productively
generated compounds. There is only one explanation for the problematic
extractions that does not falsify Russell's theory, i.e., that the fronted
constituents are not really prefixes. This, however, would imply a sys-
tematic ambiguity for all combinations of frontable prefixes with verbs:
(262)
a. Er kann immer runterkommen.
He can always down-come
He can always come down.
b. Er kann immer runter kommen.
(263)
a. Er ist hochst selten busgefahren.
he is extremely rarely bus-gone
He very seldom goes by bus.
b. Er ist hochst selten bus gefahren.
At least for the cases of prefixes derived from directional adverbs, this
ambiguity does not appear implausible. Consider the following sentence:
(264) Er kann immer weiter runter kommen.
he can always farther down come
He can always come down farther.
Most speakers would not consider runter to be a part of the verb in
(264). The coexistence of an adverb and a prefix runter is well
motivated. Similar evidence can be found for ambiguities of adjective-
verb and noun-verb combinations.
(265)
a. Der Motor muB sich warmlaufen.
the motor must self warm-run
The motor has to warm up.
b. Der Motor muB sich ganz warm laufen.
the motor must self completely warm-run
The motor has to war-m up completely.
(266)
a. Peter will jetzt kaffeetrinken.
Peter wants now coffee-drink
Peter wants to have his coffee now.
b. Peter will jetzt starken Kaffee trinken.
Peter wants now strong coffee drink
Peter wants to have a strong cup of coffee now.
For prefixes that are remnants of longer phrases, such as bus in
busfahren, the ambiguity is harder to defend. What should be the cate-
gory of bus in (263)? If it is a noun, an additional explanation is needed
for the ungrammaticality of the following examples, in which attempts
were made to modify the noun:
(267)
a. *Er ist hochst selten frUhen Bus gefahren.
he is extremely seldom early bus rode
b. *Er ist hochst selten Bus der Firma gefahren.
he is extremely seldom bus the company rode
All denominal prefixes whose source is a prepositional phrase, as in mit
dem Bus, exhibit this behavior. The same is true for all denominal
prefixes that are reduced noun phrases, if the source noun was a count
noun. The only reason that kaffeetrinken in (266a) is structurally am-
biguous is the fact that the head noun of mass nouns (as well as abstract
nouns) can occur without a determiner.
The present analysis has no difficulty in providing for the extractions
of nonseparated prefixes in (252) and (253). But those topicalizations
that are easiest for Wunderlich's and Russell's analyses appear problem-
atic for my grammar. It is not the lexical approach that rules out the
grammatical sentences in (268), (269), and (270) but rather my refusal to
commit myself to a syntactic contraction metarule that could easily be
stated. Prefix and verb are siblings of the verb's complements and do
not constitute a single constituent. Therefore, the rules proposed in
Chapter 3, Section 4 for handling topicalization will not provide for the
fronting of the complex verb as a unit, as in (250) and (251). However,
as I mentioned in the section on topicalization Chapter 3, Section 4, only
part of the story has been told. There are certain types topicalization
that are not part of the chosen fragment and that require extension of
the grammar. Among them are topicalizations of a verb, together with a
complement or adjunct.
(268) Mit dem Bus fahren kann er auf keinen Fall.
by the bus go can he in no case
Under no circumstances can he take the bus.
(269) Nach hause gehen kannst du
to home go can you
Home you can go later.
spater.
later
(270) Das Buch gefunden hatte er zwar nicht,...
the book found had he no-doubt not
Of course he had not found the book, ...
Extractions of this kind will be investigated in Chapter 6, Section 3.5. It
is obvious that whatever rules must be designed to account for the
topicalizations in (268)-(270) will also allow the fronting of prefix plus
verb, as exemplified in (252) and (253). It can be concluded, therefore,
that a solution for the fronting of complex verbs will depend on a satis-
factory analysis of all topicalizations in which the verb is fronted
together with other material.
But there is another apparent obstacle to the proposed analysis. The
presented rules not only support a fronting of the prefix but also a front-
ing of the nonfinite verb with the prefix. However, as the following ex-
amples illustrate, these topicalizations should be ruled out:
(271)
a. *Laufen sollte Peter ski (nicht springen).
run should Peter ski (not jump)
b. *Klettern wird er runter (nicht fallen).
climb will he down (not fall)
c. ??Hammern wollen wir den Stab flach (nicht
hammer want we the rod flat (not
walzen) .
roll)
All three complex verbs carry an autonomous meaning. And all three of
the verbs topicalize if they occur without prefixes:
(272)
a. Laufen sollte Peter (nicht springen) .
run should Peter (not jump)
Run is what Peter should be doing.
b. Klettern wird er (nicht fallen).
climb will he (not fall)
Climb is what he will do.
c. Hammern wollen wir den Stahl (nicht walzen).
hammer want we the steel (not roll)
We want to hammer the steel, not roll it.
There is no obvious semantic explanation for the ungrammaticality of
(271).
My proposal shares its apparent deficiency with Wunderlich's ap-
proach, since he would have to introduce an additional mechanism to
keep the verb from being fronted. This follows from the fact that the
incorporation rules are rules of optional reanalysis. Russell's grammar
does not permit generation of the problematic cases. 55
However, I would not have called the problem an apparent one, had
it been apparent to me that there really is a problem for my analysis.
The observed restriction on the fronting of nonfinite verb forms is much
more general. Neither prefixation, incorporation, nor compound
formation can be used to explain the following ungrammatical examples:
(273)
a. ??Klettern wird er weit runter.
climb will he far down
b. ??Hammern wollen wir den Stahl ganz flach.
hammer want we the steel totally flat
c. ??Gehen kann er nach hause.
go can he to home
It will require much more research to determine the conditions on the
55If he does not rule out the structural ambiguities discussed on page 102, Russell's analysis
will also have deal with this problem.
topicalization of verbs. There seem to be certain kinds of verb
complements from which the verb cannot be separated freely. Non-
specific noun phrases belong to this class.
(274)
a. Trinken will er den schwarzen Kaffee (nur nicht
drink wants he the black coffee (only not
kochen) .
cook)
He wants to drink the black coffee (he just doesn't
want to make it).
b. ??Trinken will er schwarzen Kaffee (nur nicht
drink wants he black coffee (only not
kochen) .
cook)
He wants to drink black coffee (he just doesn't
want to make it).
Often directional adverbials have a similar effect.
(275)
a. Reiten will er.
ride he will
He wants to go horseback riding.
b. ??Reiten will er zum Nachbarn.
ride wants he to neighbor
He wants to ride to the neighbor on horseback.
An interesting observation can be made by considering topicalizations
out of idioms that encompass a verb and one of its complements. Not
surprisingly, the whole idiomatic phrase can be topicalized.
(276)
a. Die Leviten lesen werden wir dem Burschen.
the Levites read will we the scoundrel
We shall teach this scoundrel a lesson.
b. Den Garaus machen werden wir dem Gesindel.
the out make will we the hoodlums
We'll kill those hoodlums.
c. Eine Abfuhr erteilen werden wir dem Aufwiegler.
a removal give will we the instigator
We'll tell the rabble-rouser- to shove off.
Separable Prefixes and Topicalization
107
d. In die Quere gekommen waren wir den Polizisten.
in the way come were we the police men
We had gotten into the policemen's way.
But it is also possible to front the idiomatic complement alone, which is
rather unexpected because the fronted phrase does not have an independ-
ent meaning.
(277)
a. Die Leviten werden wir dem Burschen lesen.
b. Den Garaus werden wir dem Gesindel machen.
c. Eine Abfuhr werden wir dem Aufwiegler erteilen.
d. In die Quere waren wir den Polizisten gekommen.
The corresponding examples of topicalizations in which only the verb IS
fronted are all ungrammatical:
(278)
a. *Lesen werden wir dem Burschen die Leviten.
b. *Machen werden wir dem Gesindel den Garaus.
c. *Erteilen werden wir dem Aufwiegler eine Abfuhr.
d. *Gekommen waren wir dem Polizisten in die Quere.
It could be shown that the fronting of nonfinite verbs by topicalization is
much more restricted than its representation in the grammar for the
fragment reflects. The inability of the grammar to prevent unacceptable
verb topicalizations does not result from of the suggested analysis of
separable prefixes, but is rather consequence of an insufficiently con-
strained account of topicalization. I shall return to this problem in
Chapter 6, Section 3.5 and shall proceed now with a brief summary of
the long section on separable prefixes.
5 Conclusion
The most predominant characteristics of my approach to separable verb
prefixes may be found in the respective roles played by lexicon and syn-
tax in the generation of complex verb constructions. In some important
ways my analysis is the reverse of alternative proposals. Whereas Wun-
derlich (and, to a certain degree, also Russell) combines prefix and verb
on the syntactic level, I do this in the lexical component. Whereas Rus-
sell (as well as traditional grammars) assumes a morphological process
that forms a compound word, I deny the existence of such a rule and
rather assume a contraction process that leaves the morphological and
syntactic structure unaffected. Instead of repeating in summarized form
the evidence I have gathered in favor of my view, let me recall an obser-
108
Separable Prefixes
vation concerning orthography that was made at the beginning of this
section. The systematic orthographic insecurity felt by native speakers
was interpreted there as an indication of missing or ambivalent intuition
about the proper word boundaries in complex verbs. It was suggested
that an adequate theory should account for this perceived uncertainty.
Wunderlich's theory provides an explanation. Optional reanalysis clearly
gives speakers a choice. However, Wunderlich's theory would also
predict improvement of the speakers' intuition in cases in which his
analysis accepts only the incorporated reading, as in busfahren,
skilaufen, and klavierspielen. In my opinion, this does not seem to be
the case, but controlled experiments are needed to verify this conjecture.
Russell's analysis does not explain the missing intuition. If one at-
tempted to construct an explanation on the basis of systematic categorial
ambiguities of prefixes, the same predictions as those generated by
Wunderlich's theory would follow.
The mixed lexical and syntactic analysis I have suggested does not
need to rely on ambiguities to explain the ambivalent intuition of
speakers. In one component of the grammar, i.e., in the lexicon, the
complex verb is simply one unit, whereas the syntax includes two en-
tities. I do not consider this strategy a language specific solution to a
language specific problem. The problem of syntactically discontinuous
lexical units comes up in many languages, among them English. Al-
though within this book I cannot elaborate on the interesting similarities
between German separable prefix verbs and a subclass of phrasal verbs in
English, it should be pointed out that the mixed lexical and syntactic
analysis offers a promising approach to English particle verbs as well.
Simpson (1982) presents English particle verbs and Warlpiri preverbs as
counterexamples to the Lexical Integrity Hypothesis Chomsky (1970),
Jackendoff (1972), Selkirk (1982). The problem for current standard
views on lexical morphology can be described as follows.
A word undergoes two levels of morphology before it enters the syn-
tax. All three levels build structure whenever they combine things. No
process can use the structure that is built on the other levels. Using
structure of a "later" stratum is simply ruled out by the iterative defini-
tion of the modular model: the structure has not been built yet. Looking
back is ruled out either by stipulation as in older versions of the Lexical
Integrity Hypothesis or by some structure changing operation that
destroys the internal structure each time before a verb transfers from one
level to the next. The Lexical Integrity Hypothesis forbids syntactic
processes to change the internal structure of lexical constituents. Move-
ments into or out of words are therefore ruled out. The Bracket Erasure
Convention, proposed by Pesetsky (1979), ensures that the internal
brackets, created by the processes of a morphological level, are erased be-
fore the word leaves this level. Any principle that effectively prevents
the syntax from changing the internal structure of words will automati-
Conclusion 109
cally account for the general observation that words cannot be separated
by syntactic processes such as left- or right-hand movements.
English particle verbs and Warlpiri preverbs are described as prob-
lematic for provisions of such kind, since their morphological constituents
may in fact move apart. The rule that lets them be separated has to be
syntactic: the interfering material does not belong to the category of the
discontinuous lexical constituent. The verb-particle compound (call up)
can be separated by a NP as in (279b).
(279)
a. Peter calls up Paul.
b. Peter calls Paul up.
Simpson's suggested solution is a relaxation of the Lexical Integrity
Hypothesis or likewise of the corresponding part of the Bracket Erasure
Principle. She proposes that the result of a verb-particle combination be
of a category yl instead of Y and that bracket erasure be restricted to
the categories A, N, P, y.
The proposal may suffice to provide the basis for movement rules
such as a stylistic rule of Particle Movement. A transformational
analysis of German would clearly benefit from Simpson's approach, since
the movement rule that fronts the finite verb in main clauses leaves the
separable prefix behind. In Bierwisch (1971), the transformation T16 for
verb fronting
56
requires the previous application of an additional
obligatory rule, T15,57 that merely inserts the appropriate word bound-
ary for the verb. However, as Simpson concedes in a footnote, 58 the
solution does not extend to non transformational phrase structure
oriented frameworks such as LFG and GPSG. There a contiguous con-
stituent cannot be separated.
The basic misconception that is to blame for the problem Simpson
attempts to solve, is that lexical processes such as verb-particle and Ger-
man prefix-verb combination generate constituent structure. The
problem disappears as soon as this misconception is abandoned. The
solution proposed here for German separable prefix verbs, on the other
hand, extends quite naturally to other frameworks as long as they have a
complex feature system and phrase structure rules. If any doubts
remain, they are concerned with the resulting power of the lexicon.
There need to be non-structure-building rules in the lexicon indepen-
dently of the phenomena discussed here. Some relation changing rules,
56Bierwisch (1965), p. III
57Bierwisch (1965), p. 109
58Simpson (1983), footnote 11, p. 284f
such as detransitivization and Dative Shift, could hardly be analyzed as
introducing a null morpheme. It would be an interesting generalization,
if in fact the only non-structure-building rules in the lexicon involve rela-
tion changes. Some of the rules needed for complex verb formation
change the subcategorization of the verb. The separable prefix ein
replaces an obligatory directional adverbial phrase.
(280)
a. Peter wird das Buch in die Tasche
Peter will the book in the bag
Peter will put the book into the bag.
b. *Peter wird das Buch stecken.
Peter will the book put
c. *Peter wird das Buch in die Tasche
Peter will the book in the bag
d. Peter wird das Buch einstecken.
Peter will the book in-put
Peter will pocket the book.
stecken.
put
einstecken.
in-put
However, it was pointed out in Section 3 on page 94 that, contrary to
Wunderlich's hypothesis, not all prefix-verb combinations are relation-
changing if the standard views of subcategorization are assumed. It is to
hope that future research on sub categorization and on the mapping from
thematic roles to syntactic entities will give rise to a more general ap-
proach to sub categorization, i.e., one that treats traditional sub-
categorization and the mapping from thematic roles to free adjuncts as
two instances of the same concept. It will be in this wider sense that
Wunderlich's hypothesis can support the generalization that all non-
structure-building rules are relation-changing.
For the formal power of the framework there is no difference be-
tween the two views, for both, subcategorization information and the
syntactic information about the nonsubcategorized-for prefix or particle,
have to be encoded in the lexical category. The only way to encode in-
formation in a complex category is by means of features. The properties
of the feature set are crucial for the formal properties of the framework.
Infinite feature sets or infinite feature value sets lead to infinite category
sets. If German separable prefixes or English verb particles constitute an
open class and if all separable prefixes correspond to features, or values
of a feature PREFIX, then it appears that the set of nonterminals will
indeed be nonfinite. Now it happens that in both German and English,
there is no productive process that creates separable prefixes or verb par-
ticles from open word classes such as verbs, nouns, and adjectives. The
only productive process of this kind in German combin=s verbs with an
extensionally specifiable class of local adverbs. It can be concluded that
the proposal of a non-structure-building lexical solution for the genera-
Conclusion
111
tion of potentially discontinuous lexical units does not create infinite eat-
egory sets. It is completely based on already existing formalisms.
In addition, it needs to be noted that any truly productive process
such as prefix formation from local adverbs, could be expressed in a more
sophisticated feature system as it is used in most versions of GPSG with-
out making the prefix itself the value of the feature PREFIX. In a sys-
tem with category-valued features, i.e., features that can have categories
as values, the feature bundle that specifies a range of potential prefixes
can be used as a value of the prefix feature on the verb. Actually, those
verb entries in which the lexical string of the prefix is encoded as t:1e
value of the PREFIX feature are then a special case of verb entries in
which a class of possible prefixes is encoded in a category valued feature.
Recent developments in GPSG treat underspecified and fully specified
categories on a par. A category that bears no information except a lex-
ical string feature is just an impoverished category. The lack of further
syntactic features correlates here with an impoverished meaning, for it is
exactly the completely nonproductive cases, in which the prefix does not
carry any independent semantics. In such a theory, a category valued
prefix feature is just an instance of some broader class of features that
are needed to describe incorporation phenomena across languages. For
GPSG, this is an important step towards a richer theory of the lexicon.
Let me summarize. The investigation of the notoriously difficult
problem of German separable-prefix verbs has produced a number of
relevant results.
Prefix-verb combinations are lexical units.
Some prefix-verb combinations are fully lexicalized and
semantically noncom positional, others are the result of
productive rules. Most lie on the scale between the extremes.
This is only more evidence against a strict boundary between
word formation rules and productive lexical rules.
Every prefix fills a thematic role slot. Sometimes the role
slot corresponds to a syntactic argument position under the
standard interpretation of subcategorization.
There is strong evidence against viewing prefix and verb as
one syntactic constituent.
The framework of GPSG provides the tools for treating prefix
and verb as one lexical entity but distinct syntactic con-
stituents.
The lexical processes that combine prefix and verb require a
more developed lexical component of GPSG, as it is indepen-
dently warranted.
The approach does not increase the generative power of the
formalism.
At least for German, the mixed lexical and syntactic ap-
proach provides a solution to an apparent contradiction be-
tween lexical integrity and syntactic discontinuity.
The approach is likely to extend to similar problems in other
languages and probably also to other frameworks.
Chapter 5
The Order of Verbal
Complements and Adjuncts
1 Introduction
The grammar rules introduced so far do not influence the order of the
syntactic arguments of the verb and free adjuncts where these form a
contiguous string.
59
In Chapter 1, Section 4, it was pointed out that all
six permutations of subject, indirect object, and direct object in
(45a)-here repeated as (281)-give rise to grammatical sentences in
German.
(281) Dann wird der Doktor dem Patienten die Pille geben.
then will the doctor the patient the pill given
Then the doctor will give the patient the pill.
Since there are no LP rules that could rule out any of the six possib.e
sequences, all six grammatical sentences can be generated. Unfor-
tunately, the grammar allows just as easily the generation of (46a),
which is repeated here as (282).
(282) ??Dann hatte einen groBen Spielzeuglaster ihm es gegeben.
Then had a big toy truck him it given.
Then it had given him a big toy truck,
In Chapter 1, Section 4, a number of principles were described that
determine the acceptability associated with permutations of subject, ob-
59Sometimes I use the term complement for all syntactic arguments of the verb including
subject, objects, to-infinitivals, etc. The term adjunct is used for adverbial phrases ar.d
other modifiers of verb phrases or clauses.
113
114 The Order of Verbal Complements and Adjuncts
ject, and indirect object. Some of those principles extend to other
complements and adjuncts. The selected principles that I will be con-
cerned with in this section are repeated below under (283).60
(283) Focus follows nonfocus
The unmarked order is SUBJ, IOBJ, DOBJ.
Personal pronouns precede other NPs.
The interaction of these principles is reponsible for the contrast in
grammaticality between (281) and (282). In this section I will incor-
porate the principles into the grammar for the framework. To enable
them to interact appropriately, the framework needs to be modified by a
redefinition of its LP component. The redefinition will be explained and
its effects demonstrated. The last subsection evaluates the theoretical
implications of the resulting integration of syntactic and pragmatic con-
cepts within the grammar.
As already indicated, the analysis of the order among the syntactic
arguments of the verb is based on the assumption that nonsyntactic and
syntactic ordering principles are not separated into different components
of the grammar. The part of the metagrammar that controls the linear
sequence of constituents is the LP component. The following LP rules
have been introduced in previous sections:
(284)
a. V < X
+MC
b. X < V
-MC
c. X2 < SEPREF
d. +TDP < X
One could add a set of LP rules (285) that correspond to the ordering
principles under (283):
(285)
a. +NOM
<
+DAT
b. +NOM
<
+ACC
c. +DAT
<
+ACC
d. -FOCUS
<
+FOCUS
e. +PRONOUN
<
-PRONOUN
60See Section 3.1 for some consequences of this selection.
In trod uction
115
A feature FOCUS has been added here that designates a focused
constituent. I will discuss this feature in Section 3.
However, note that the new LP rules do not resolve the problem of
ordering-principle conflicts, for the violation of one LP rule is enough to
rule out an ordering. The LP rules would indicate the impossibility of a
sentence having a pronominalized object NP next to a nonpronominal
subject NP, as in (28Ga) or (28Gb), since either sequence would violate
one LP rule:
(286)
a. Dann wird der Doktor ihn sehen.
then will the doctor him see
Then the doctor will see him.
b. Dann wird ihn der Doktor sehen.
then will him the doctor see
The order in (28Ga) violates LP rule (285e), since the pronoun follows
the full NP; the order in (28Gb) violates (285b), since it deviates from the
unmarked order. On the other hand, the absence of the new LP rules
would incorrectly predict that always all permutations are acceptable.
The next section introduces a redefinition of LP rules that provides a
remedy for this deficiency.
2 The Modified Framework
Before introducing a new definition of LP rules, let me suggest another
modification that will simplify things somewhat. The LP rules con-
sidered so far are not really such in the sense in which they were defined
by their originators. After all, LP rules are defined as members of a par-
tial ordering on VN U VT. Our rules are schemata for LP rules at best,
abbreviating the huge set of LP rules that are instantiations of these
schemata. This definition is an unfortunate one in several respects. It
not only creates an unnecessarily large set of rules (VN contains
thousands of fully instantiated complex symbols), but also suppresses
some of the important generalizations about the language. Clearly, one
could extract the relevant generalizations even from a fully expanded LP
relation-e.g., conclude that there is no LP rule whose first element has
-MC and its second element NP. However, it should not be necessary to
extract generalizations from the grammar; it should express these
generalizations directly. Another disadvantage follows from the choice of
a procedure for arriving at the fully expanded LP relation. For instance,
should all extensions that are compatible be instantiations of (284) LP
rules? If so, then (311) is an instantiation of (304a):
(287)+MC NP
+DEF
<
+FIN
Yet nothing can be a verb and definite simultaneously, nor can NPs
be finite. (287) is a vacuous rule. Whether it is a LP rule at all will
depend on the way the nonterminal vocabulary of the object grammar is
defined. If it includes only the nonterminals that actually occur in rules,
(287) is not an LP rule. In this case we would need a component of the
metagrammar, the feature instantiation principles, to determine another
component of the metagrammar, the LP component. 51 LP will be
redefined as a partial order on 2
F
, where F is the set of syntactic
features.
52
The second and more important change can best be described by
viewing the LP component as a function from a pair of symbols (which
can be characterized as feature sets) to truth-values, telling us for every
pair of symbols whether the first can precede the second in a linearized
rule. Given the LP relation {(aI' {31),(a
2
, {32), ... ,(a
n
, {3n) and a pair of
complex symbols (I, 8), the function can be expressed as follows:
(288) C
1
/\ C
z
/\ ... /\ c
n
where
c
1
= --. (0'15;8 /\ {315;"{)
for 1< i < n
- -
We call the conjunct clauses LP conditions; the whole conjunction is
a complex LP condition. The complex LP condition allows "t to precede 8
on the right-hand side of a CF-PS rule if every LP condition is true. An
LP condition c
1
derived from the LP rule (a., {3.) is true if it is not the
I I
case that "t has the features {3. and 8 has the features o.. Thus, the LP
I I
rule NP < VP stands for the following member of the LP relation
({+N, -V, +2BAR},{-N, +V, +2BAR}). The LP condition following
from this rule prevents a superset of {-N, +V, +2BAR} from preceding
a superset of {-N, +V, +2BAR} , i.e., a VP from preceding an NP.
But it should be noted that nothing prevents us from writing a fic-
titious LP rule such as
51The widely used notation for noninstantiated LP rules and the feature instantiation
principles could be regarded as meta-metagrammatical devices that inductively define a
part of the metagrammar.
62Remember that, in an X -sy ntax, syntactic categories abbreviate feature sets
NP={ +N,- V,+2BAR}. The definition can easily be extended to work on feature graphs
instead of feature sets.
The Modified Framework 117
(289) +PRONOUN < -ACC
German has verbs like Zehren that take two accusative noun phrases as
complements. If (289) were an LP rule, the resulting LP condition
defined as in (288) would rule out any occurrence of two pronominalized
sister NPs because either order would be rejected.
53
It is an empirical
question whether one might ever find it useful to write LP rules as in
(289), i.e., rules a < {3, where a U {3 could be a subset of a complex
symbol. Let me introduce a minor redefinition of the interpretation of
LP that will take care of such cases as (289) and, at the same time pave
the way for a more substantial modification of LP rules. LP shall again
be interpreted as a function from pairs of feature sets (associated with
complex symbols) to truth-values. Given the LP relation
{(a
1
, {31),(a
2
, {32),,(a
n
, {3n)} and a pair of complex symbols h8), the
function can be expressed as in (290):
(290) c
1
/\ c
2
, /\ ... /\ c
n
where
c
1
= (0'15;8 /\ {315;,,{)-+(a
1
5;"{ /\ {315;8)
for 1::; i ::;n
That means "t can precede 8 if all LP conditions are true. For instance,
the LP condition of LP rule (289) will yield false only if "t is +ACC and 8
is +PRONOUN, and if either "{is -PRONOUN or 8 is -ACC (or both).
Now let us assume that, in addition to the kind of simple LP rules
just introduced, we can also have complex LP rules consisting of several
simple LP rules and notated in braces as in (291):
(291)
{
+NOM
+NOM
+DAT
-FOCUS
+PRONOUN
<
<
<
<
<
+DAT }
+ACC
+ACC
+FOCUS
-PRONOUN
The LP condition associated with such a complex LP rule shall be the
disjunction of the LP conditions assigned to its members. LP rules can
be generally defined as sets of ordered pairs of feature sets
{(al'{31),(a
2
,{32), ... ,(a
m
,{3m)}' which are either notated with braces as in
(10), or, in the case of singletons, as LP rules of the familiar kind. The
members of such a set are called atomic LP rules.
53In principle, there is nothing in the original ID/LP definition either that would prevent
the grammar writer from abbreviating a set of LP rules by using (289). It is not quite
clear, however, which set of LP rules is thereby abbreviated.
If a complex LP rule {(al'/31),(a
2
,/32), ... ,(a
m
,/3m)} is applied to an or-
dered pair of feature sets ("1,8), i.e., if it is used to determine whether the
two complex symbols "I and 8 can occur in this order on the right-hand
side of a rule, then the complex LP rule is interpreted as a LP condition
of the following form:
((0'1 5; 8/\/3
1
5; "() V (0'2 5; 8/\/3
2
5;"{) V V(a
m
5; 8/\/3
m
5; "())
-+ ((0'1 5;"{ /\ /3
1
5;8) V (0'25; "( /\ /3
2
5; 8) V V (am 5; "( /\ /3
m
5; 8)).
Thus, any of the atomic LP rules within the complex LP rule can be
violated as long as the violations are sanctioned by at least one of the
atomic LP rules. An atomic LP rule sanctions a violation if the rule
would be violated by the reversed order (8,"{).
Notice that, with respect to this definition, II regular II LP rules, i.e.,
singletons, can be regarded as a special case of complex LP rules.
I would like to suggest that the LP rules in (284) and (291) are a
subset of the LP rules of German. This analysis makes a number of em-
pirical predictions. For example, it predicts that (292a) and (292b) are
grammatical, but not (292c):
(292)
a. Dann will der Doktor dem Mann die Pille geben.
-FOCUS +FOCUS -FOCUS
+NOM +DAT +ACC
then wants the doctor the man the pill give
b. Dann will der Doktor die Pille dem Mann geben.
-FOCUS -FOCUS +FOCUS
+NOM +ACC +DAT
then wants the doctor the pill the man give
c. ??Dann will der Doktor die Pille dem Mann geben.
+NOM +ACC +DAT
then will the doctor the pill the man give
In (292c) the atomic LP rules +DAT < +ACC and
-FOCUS < +FOCUS are violated. No other atomic LP rule legitimizes
these violations and therefore the sentence is bad. It can be seen already
that the concept of grammatica.lity implicit in this analysis goes beyond a
simple mapping from strings to truth-values, since features based on dis-
course roles can be deduced only from the interaction of context and in-
tonation. Sentences (292b) and (292c) are represented by the same
string. However, the rule that is needed to generate the latter sentence
will not be derived.
This agrees with the findings of Lenerz (1977), who tested a large
number of sample sentences to determine the interaction of the un-
marked syntactic order and the ordering preferences introduced by dis-
The Modified Framework
119
course roles. So far nothing has been said about the wide range of
degrees of acceptability that must be dealt with in regard to ordering
violations. Although I consider this justified as a means of preventing
the grammar from generating (292c), I am aware that even in a context
that forces the NP die Pille to denote the focus, the sentence is still more
grammatical than, for instance, (293), in which the subject and the finite
verb have been permuted. I assume the feature assignment of (292b).
(293) Dann der Doktor Wird die Pille dem Patienten geben.
But notice that no claims have yet been made about the degree to which
the atomic LP rules contribute to the grammaticality of an order. This
question will be one of the issues discussed in Section 3.1.
3 Discussion of the Analysis
Proposals of two different types need to be discussed: modifications of
the framework; analysis of noun phrase order within the fragment.
Let me start with the analysis. The suggested complex LP rule (291)
uses syntactic features to state ordering constraints. The least controver-
sial feature in this connection is PRN. A syntactic feature that marks
personal pronouns is needed for other purposes besides regulating linear
order. To mention just one application: the range of postnomina.l
modifiers is much more restricted for personal pronouns than for non-
pronominal noun phrases. However, there is one potential argument
against my use of this feature. One might attempt to reduce the number
of ordering principles by deriving some of them as special instant.iations
of others. Such a reductionist approach could suggest that the ordering
principle positioning personal pronouns before other noun phrases, must
be subsumed under the principle that refers to the focus of the sentence.
After all, personal pronouns refer to mutually known entities and it is
mutual knowledge of a referent that is also important for the topic-focus
structure of the sentence. Pronouns tend to precede other noun phrases,
while the topic tends to precede the focus. Although I suspect the exist-
ence of a strong correlation between the principles and can well imagine
some historic causal connection, I am nevertheless not ready to derive
one principle from the other.
The focus of a sentence does not have to refer to entities that are not
mutually known, not even to entities that are new in the current context.
Imagine the following fragment of a dialogue. The two sentences are
spoken by the same person:
(294) Und weiBt du, wem von den beiden der Schaffner
and know you, whom of the both the conductor
beim Aussteigen geholfen hat?
at getting-off helped has
Beim Aussteigen hat der Stoffel ihm geholfen.
at getting-off has the booby? him helped
And which of the two, do you think, did the the conductor help
getting off? He helped him to get off.
The noun-phrase ihrn in the second sentence clearly represents the focus
in the sentence.
The feature +PRN also influences the order among noun-phrases
that do not differ in their FOCUS feature assignment. It is easy to find
a context for (295) and (296) in which only the NP ein Buch is in the
focus:
(295) Gestern hatte er seinem Freund ein Buch geliehen.
yesterday had he his friend a book loaned
Yesterday he had loaned a book to his friend.
(296) Gestern hatte ihm sein Freund ein Buch geliehen.
yesterday had him his friend a book loaned
Yesterday his friend had loaned him a book.
In the case of the feature PRN, the reductionist approach has to be
rejected. Similar considerations have been under debate about the role of
other features such as definiteness for the ordering of noun-phrases. I
have nothing more to say about this question, except that it is, of course,
a worthwhile attempt, if it can lead to a reduction of the number of or-
dering principles.
In the complex LP rule, the principles of unmarked order are stated
with reference to case marking and case features, just as all other syntac-
tic features included within category specifications. However, in (4) these
same principles were introduced in terms of grammatical functions, such
as subject and object.
The theoretical question whether the unmarked order needs to be
determined on the basis of grammatical functions or categories is quite
an interesting one, since most proponents of GPSG reject the role gram-
matical functions play in other frameworks, such as Lexical Functional
Grammar and Relational Grammar. Within our limited framework, no
evidence can be found for or against the inclusion of grammatical func-
tions within the set of syntactic primitives. This is so because grammati-
cal functions and case marking in the fragment are isomorphic. In Chap-
ter 5, I shall discuss extensions of the fragment that destroy the isomor-
phism. Among the relevant constructions are sentential subjects, double
accusative verbs, predicatives, and genitive NPs as complements and as
possessive NP modifiers.
Discussion of the Analysis
Another feature employed by LP rule (291) is FOCUS. Despite its
name, FOCUS is a syntactic feature that is justified by syntactic facts,
such as its influence on word order. However, this syntactic feature
needs to be linked to the appropriate discourse information. The place
to do this is in the rule extension component, where features are instan-
tiated and semantic translations added to ID rules.
It is assumed that in so doing the translation part of the rules will
have to be extended anyway so as to incorporate non-truth-conditional
aspects of the meaning. For example, the full translation could be an
ordered pair of both truth-conditional and non-truth-conditional content,
extending Karttunen and Peters's (1979)54 treatment of conventional
implicature or a function from discourse situations to the appropriate
truth-conditional meaning in the spirit of Barwise and Perry (1981). The
analysis here is not concerned with choosing a formalism for an extended
semantic component, but rather with demonstrating just where the syn-
tax has to provide for those elements of discourse information that in-
fluence the syntactic structure directly.
3.1 Discussion of the General Approach
A difficult challenge confronts me in this section. The multitude of
problems connected with the regularities that govern the partially free
order of verb complements and adjuncts certainly cannot be solved in
this book, moreover, it cannot even be addressed in its entirety. On the
other hand, I would like to convince the reader that, by using and
modifying the formalism of GPSG, I have created a useful tool for fur-
ther research and grammatical description. A contradiction of this sort
is usually solved by applying the formalism successfully to a fragment of
the set of problematic constructions. The choice and size of the fragment
will determine the degree to which the reader assumes by extrapolation
that the approach can be extended to cover more or even all of the
relevant data. That strategy is followed throughout this chapter.
In the case of the current problem there exists a major obstacle to
the standard strategy. It is virtually impossible to find a fragment of
data that does not exhibit far more ordering regularities than could pos-
sibly be considered in this section. Lenerz (1977) had to deal with this
problem. The best way to investigate mutual dependencies among a set,
of variables is to set up conditions that keep all but two variables con-
stant during the experiment. When Lenerz explored the existence of an
unmarked order, he tried to keep all ordering principles constant with
the exception of unmarked order and focus assignment. Not only did he
64To be more precise, Karttunen and Peters actually make their translations ordered triples
of truth-conditional content, implicatures, and an inheritance expression that plays a role
in handling the projection problem for presuppositions.
have to ensure that the permuted phrases agreed in phonological
heaviness, the choice of determiners, pronominalization, and metrical
properties; he also had to rule out orderings that are motivated by em-
phatic or contrastive stress. There are many cases in which both prin-
ciples, unmarked order and nonfocus-focus, can be overridden by contras-
tive or emphatic stress assignment. As hard as it sometimes is to tune
informants to an assignment of discourse roles to constituents, it is
equally hard to rule out certain classes of contexts for the interpretation.
In fact, it is much harder, since positive context conditions can be in-
voked by instantiation through an example, whereas it is much more dif-
ficult to rule out whole classes of contexts by instantiating a specific con-
text. It is also hard to locate the borderline between focus and emphatic
or contrastive focus-if such a borderline actually exists.
One of the arguments presented in favor of separating syntactic and
stylistic rules-sometimes into two different components-is the dif-
ference in degree of ungrammaticality that results from the violation of
such rules. The contrast between (292c) and (293) would come under the
scope of this argument. Obviously, I cannot agree with the argument as
it stands. Even if I do not contest differences between pragmatically
based and other kinds of rules, my proposal still calls for formulating all
ordering rules as syntactic rules in the same component of the metagram-
mar. Consider the following English examples:
(297)
a. Peter is taking the car to the shop.
b. *Peter took his car the shop to.
c. *Took his car to the shop Peter.
d. *Peter is taking to the shop his car.
Sentences (297b), (297c), and (297d) are ungrammatical ordering varia-
tions of (297a). All permutations involve sibling constituents only.
Clearly, (297b) is worse than (297d). Example (297c) lies somewhere be-
tween (297a) and (297c). An explanation for the contrast could be con-
structed by claiming that the less grammatical sentences are all inter-
preted in consequence of an illegal application of a stylistic rule. In the
case of (297d) the rule of Heavy-NP-Shift could have been applied to a
"light" NP. Not only would this explanation fail to account for the con-
trast of (297b) and (297c); it is also unnecessarily complicated.
Let me offer some ideas about the interpretation of the formalism I
have introduced. Linear order is a syntactic phenomenon. All ordering
regularities need to be encoded in syntactic rules. LP rules, just like
other rules of the metagrammar, can be associated with certain types or
modes of language use-e.g., dialogues, speeches, letters, and technical
wntmg. Some rules are more limited by such associations than others.
An example of a rule that is very limited is left dislocation. Competing
ordering principles can be weighted according to the type or mode of
communication. Clear differences in ordering preferences between
monologues and dialogues have been detected by Jecklin (1973). There
are also weight differences that depend on speaker idiosyncracies. Most
likely there are also absolute weight differences among ordering pinciples;
for instance, the principle that orders pronouns before nonpronominal
NPs seems stronger throughout than the heaviness principle.
It is an open question whether scalar features are needed to express
quantitative influences such as the length difference of two constituents
or degrees of emphasis. My conjecture is that features with discrete
finite value sets are sufficient. Quantitative influences such as degree of
emphasis and heaviness proportions can still trigger the choice of one 01'-
dering principle over another in the processing model.
The proposed formalism suggests the following criterion for distin-
guishing between stylistic and syntactic acceptability. As long as one or-
dering principle licenses an order, it is syntactically well-formed. The de-
gree of markedness increases with the number or total weight of violated
principles.
Although phonologically and pragmatically based principles are also
encoded in syntactic rules, the difference among components is not lost.
One of the reasons that (292c), here repeated as (298), does not seem 11;3
ungrammatical as based ordering violations that are purely syntactically
based is that the hearer still has two possible interpretations left, even if
they do not fit the context.
(298) ??Dann will der Doktor die Pille dem Mann geben.
+NOM +ACC +DAT
then wants the doctor the pill the man give
Focus could be on dem Mann or there could be emphatic or contrastive
focus on die Pille. The error could thus be perceived as a misinterpreta-
tion of the context or-if the context does not rule out the emphatic
reading completely-the deviation could be attributed to mispronouncia-
tion.
The claims put forward in the last paragraph will undoubtedly sound
familiar to anyone who has read or thought about the problems of
markedness and stylistics in connection with free word order. The
novelty of my approach lies merely in the existence of a formal com-
petence model
55
that can be used in connection with an appropriate
model of performance to test descriptive accounts involving the inter-
action of syntactic, pragmatic, and phonological regularities. For a num-
55A complete competence model of linear ordering might require some basic or initial values
for the weights assigned to atomic LP rules.
ber of reasons, valid tests of this kind need to be large scale, controlled
experiments. The following must be considered:
The large number of interacting factors
The quantitative nature of rule weighting
The quantitative influences exerted by such scalar entities as
emphasis or heaviness
The scalar nature of acceptability.
I would like to conclude this section by illustrating different degrees
of acceptability by stepwise modification of the ungrammatical example
(282), which I repeat here as (299).
(299) ??Dann hatte einen groBen Spielzeuglaster ihm es gegeben.
+ACC +DAT +NOM
Then had a big toy truck
Then it had given him a big toy truck.
him it given
Note that under the present analysis, no feature assignment can make
(299) grammatical, because no assignment of FOCUS can legitimize all
three violations of unmarked order. In addition to the principles of un-
marked order, the principle that attempts to front pronouns and the
heaviness principle have been violated. Let me demonstrate a gradual
increase of acceptability through a decrease of the number of ordering
principle violations. For this purpose I shall list the ordering principles
again that are atomic LP rules in the complex LP rule (291):
(300) i. +NOM
<
+DAT
ii. +NOM
<
+ACC
iii. +DAT
<
+ACC
iv. -FOC
<
+FOC
v. +PRN
<
-PRN
The first sentence is (301), a variation of (299), in which the heavy object
noun-phrase has been replaced by a shorter NP.
(301) Dann hatte ein Buch ihm es gegeben.
+ACC +DAT +NOM
then had a book him it given
Then it had given him a book.
Let us assume a probable context in which ein Buch is +FOCUS, and
the other two NPS are -FOCUS. Then the following principles are vio-
lated: i., ii., iii., iv. (twice), v. (twice).
Discussion of the Analysis 125
Next, the subject pronoun will be replaced by a proper name:
(302) Dann hatte einen Ball ihm Peter gegeben.
+ACC +DAT +NOM
then had a ball him Peter given
Then Peter had given him a Ball
Violated principles are: i., ii., iii., iv. (twice), v. Again, the sentence is
slightly better than the last one. If the direct object becomes -FOCUS
and the subject +FOCUS, the sentence improves considerably. The dif-
ferent discourse role assignment is supported by a shift in definiteness.
(303) Dann hatte den Ball ihm ein Freund gegeben.
+ACC +DAT +NOM
then had the ball him a friend given
Then the ball was given to him by a friend.
Violated Principles are: i., ii., iii., v. The following variation, finally,
should fully satisfy any teacher of German:
(304) Dann hatte ihm den Ball ein Freund gegeben.
+DAT +ACC +NOM
then had him the ball a friend given
Only two principles are violated: i. and ii.
There remains at least one more possible objection to the framework
suggested for dealing with partially free word order. Descriptive gram-
mars of German have pointed out that the unmarked order is used to
disambiguate in cases in which the grammatical functions (or argument
roles) of the noun phrases cannot be deduced from the inflectional
morphology. Sentence (135) in Chapter 3, Section 2.4, was presented as
an example of this phenomenon. Let me repeat the example:
(305) Dann hat das Nilpferd das Krokodil
then has the hippo the crocodile
Then the hippo bit the crocodile.
gebissen.
bitten
In Chapter 3, Section 2.4, I already indicated that I do not consider this
phenomenon to be a part of the grammar. My first reason is based on
examples in which the pattern is violated. In such cases the speaker is
confident that some information other than case marking will enable the
hearer to disambiguate the sententence. In (306) this other information
is the content of the sentence:
(306) Dann hatte das restliche Mobiliar das Finanzamt
Then had the remaining furniture the IRS
gepfiindet.
confiscated
Then the remaining furniture had been confiscated by the IRS.
One of the two readings would not be considered under all but the most
exceptional circumstances, i.e., the one according to which the furniture
confiscates the tax authority.
The disambiguating information can also be part of the syntactic
structure of preceding clauses. Consider the following conjunction:
(307) Erst hatte den Jungen eine Tarantel gestochen,
first had the boy a tarantula stung
und dann hatte das Madchen ein Krokodil gebissen.
and then had the girl a crocodile bitten
First the boy had been stung by a tarantula, and then the girl
had been bitten by a crocodile.
The conjunction sounds much better than the second conjunct in isola-
tion (under the slightly more sensible interpretation).
An analysis that in which the weight of the principles depends on the
degree to which ambiguities can arise would not only be extremely messy
to implement; it would also lead to the loss of an important generaliza-
tion. Many cases have been found across languages in which optional
rules can be blocked or enforced when ambiguities must be avoided. The
prevented ambiguities are of a quite different nature. In the case of
English extrapositions from subjects, for instance, there are attachment
ambiguities. It is difficult to view the following extraposition as modify-
ing the subject:
(308) The tenants hated their partying neighbors who were
sensitive to noise.
But now consider the following examples which contain additional infor-
mation that assists the hearer to disambiguate:
(309) Only those tenants hated their partying neighbors who
were sensitive to noise.
(310) Only those tenants hated the party who were sensitive to
noise.
127
Instead of making individual rules, such as the LP rules ordering noun-
phrases or the rule of relative clause extraposition, sensitive to all the
factors that can introduce or exclude ambiguities, one should rather as-
sume the existence of a universal communicative strategy to cooperate in
the avoidance ambiguities. Even if sentences formed in violation of this
strategy might at times be as unacceptable as sentences that are syntac-
tically ungrammatical, there is no evidence for the violation of gram-
matical rules.
A note on the relationship between the scale of word order freedom
across languages and the redefined LP component shall conclude this
chapter. Although all proposed LP rules of German are of the same for-
mal type they sometimes look quite different. Consider the following
two rules.
(311)
a. V < X
+MC
b.
rOOM
<
+DAT }
+NOM
<
+ACC
+DAT
<
+ACC
-FOCUS
<
+FOCUS
+PRONOUN
<
-PRONOUN
The (311a) is a singleton LP rule, (311b) is a complex LP rule. In
the unmodified ID/LP version of GPSG the difference between more and
less configurational languages is reflected in the number of LP rules. A
language with more ordering freedom will not impose as many LP rules
on its ID rules as a language with a stricter word order. The smaller
number of LP in a free word order language is not compensated by a
greater number of ID rules. In fact, the opposite might be true in many
cases: since free word order languages tend to exhibit a flatter con-
stitituent structure than English, fewer ID might be needed. On the
average, the few ID rules will have longer right hand sides. Therefore
the smaller number of ID rules will be compensated by a higher degree 0::
complexity of these rules. The smaller number of LP rules is not corn-
pensated in the syntax. It seems then that languages with free word or-
der have a less complex syntax.
This for different in the modified framework. Here, languages with
relatively strict word order require many simple LP rules; languages with
relatively free word order require fewer LP rules which are more complex
on the average. As one might expect, the position of German is some-
where in the middle. The redefined LP component appears more
plausible than an LP component that is strong in some languages and
almost nonexistent in others.
Chapter 6
Evaluation and Expendability
Of the Grammar
1 Introduction
The preceding chapter presented a grammar for a fragment of German.
The accompanying argumentation for controversial or potentially con-
troversial solutions was based on empirical evidence. The constructions
included in the fragment exemplify the relevant constituent-order
phenomena of the language, which were introduced in Chapter 1. The
grammar was written in a slightly modified version of the formally con-
strained framework of GPSG, whose virtues and design features were
described in Chapter 2. The motivation for applying this framework to
word order related phenomena in German was explained in the Introduc-
tion. On the basis of certain premises-especially permutational
variation-I intended to explore the appropriateness of a constrained,
well-defined, phrase-structure-based formalism for the description of a
language that exhibits a higher degree of word order freedom than does
the language that served as the testbed for the development of this for-
malism.
An empirically adequate grammar for a fragment of German is an
important part of an answer to the question, yet no more than a part.
Two additional questions need to be answered:
1. Has the grammar the desired properties that make it an
appropriate grammar?
and, if this is indeed the case,
2. Can the grammar be extended to constructions outside the
fragment without losing those properties?
128
Introduction 12~1
Obviously, these two questions are closely related. It is not only
abstract properties such as perspicuity or generality that need to be
preserved when the grammar is enlarged, but also the individual linguis-
tic strategies, solutions, and generalizations that gave the grammar those
properties.
In this chapter an attempt is made to answer the two questions for
the grammar that was presented earlier. The acronym GFG will be fre-
quently used to refer to the grammar of the fragment of German. The
following section will outline the relevant features of GFG, including cer-
tain linguistic hypotheses that need to be tested when the grammar is
extended. The grammar will be evaluated on the basis of these features.
In Section 3, the expandability of GFG will be explored in connection
with a variety of problematic phenomena in German syntax. The discus-
sion will focus upon the relevance of potential extensions for specific
properties of GFG, such as its primitive concepts, generative capacity:
and its basic assumptions about German constituent structure. The finaJ
section of the chapter will summarize the predicted consequences of the
extensions being considered. Directions will be suggested for future
modifications of the framework.
2 Properties of the Grammar
2.1 Basic Theoretical Properties of GFG
The grammar for the fragment inherits most features that are common
to other grammars written in the framework. A small metagrammar
with very little redundancy inductively defines a huge CF-PS grammar.
In Chapters 3, 4, and 5, it was demonstrated how the different rule sys-
tems of the metagrammar interact to create a multitude of object gram-
mar rules from a single basic rule. It was also shown at several places
within Chapter 3 that and how GFG is compatible with the current
GPSG approach to semantics. This means that the syntactic information
of an ID rule is sufficient to deduce the corresponding semantic rule(s)
that can then be associated with the corresponding CF-PS rule. The
rationale behind the use of X-syntax is the reduction of redundancy in
the system of phrase structure rules. Among other advantages, the sys-
tem of X-syntax allows a concise characterization and encoding of the
notion head. Gazdar and Pullum (1982b) have demonstrated how the
IDjLP format can operate on this notation to express generalizations
about the ordering of heads with respect to their complements. Certain
well-known observations of typologists can be easily encoded.
55
55See Stucky (1982) for a brief survey of the potentials and limitations of the IDiLP
formalism with respect to the notation of typlogical universals.
130 Evaluation and Expendability of the Grammar
There are many languages in which the position of the verb in the
sentence coincides with the position of the phrasal heads within their
constituents. Examples are Japanese and Breton. The verb-phrase and
clause rules in GFG show that in German the head can either precede or
follow its sibling constituents. Clearly, the fragment does not include
enough phrase types to make this an interesting observation about Ger-
man constituent heads in general. However, the inconsistency seems to
extend to wider segments of the language. German has both prepositions
and postpositions. As in English, not all genitive nominal modifiers fol-
low the noun. A similar split exists for other pre- and postnominal
modifiers: adjectives and participal phrases precede the noun, relative
clauses and prepositional phrases follow it. Yet one generalization can be
noted: the head always either precedes or follows all of its sibling con-
stituents. This is true for both complements and modifiers. The second
position of finite verbs in main clauses is not a counterexample. Quite
the opposite is true: in the GFG analysis, all three verb patterns follow
from instantiations of the principle that makes the head always either
the first or last constituent of the immediately dominating node.
In Chapter 2, Section 2, the choice of syntactic primitives was men-
tioned. It was pointed out that most proponents of GPSG do not con-
sider grammatical functions to be primitive concepts of the theory. GFG
does not employ grammatical functions. It relies on the system of case
marking for the mapping from syntactic categories to semantic argument
pOSItIOns. In Section 3.1.1, some problems will be discussed that arise
when the grammar is extended to verb complementations that do not al-
Iowa straight forward mapping from categorial information to the posi-
tion of the constituent in the syntactic structure.
It needs to be emphasized that the feature FOCUS does not
represent a grammatical function, as it is the case in LFG (Bresnan
(1982)). FOCUS is just a regular syntactic feature. It needs to be
linked to the non-truthconditional aspects of meaning in the rule exten-
sion component.
2.2 Generative Capacity and Implementability
One of the most difficult problems in connection with word order
freedom is to preserve constraints on generative capacity. The
framework assumed by Stucky (1981) in her GPS grammar for Makua
has Turing machine power. Although the constructions of Makua that
are discussed by Stucky do not require such a degree of generative power,
the excessive power comes in through the interaction of essential vari-
ables and multiple applications of metarules that is described in
Uszkoreit and Peters (1983).57 GFG does not employ any formalisms that
57The question of generative power was discussed in Chapter 2, Section 5.
Properties of the Grammar 131
might raise the formal power of the framework. The constraint of Finite
Closure is not violated: no metarule needs to apply more than once in the
derivation of a CF-PS rule. In addition, the set of categories is kept
finite.
The proposed extension to the LP component does not increase the
generative power either, since it does not give rise to nonfinite object
grammars. This claim is easily verified. Let us assume that the
framework is context-free before the modification, i.e., that tne
metagrammar generates a finite set of CF-PS rules. An empty LP
component-that is, a linearization component without LP rules-would
for any fully instantiated ID triple (n; a -+ ; [2) generate all CF-PS
rules (n; a -+ 7/J; [2), where 7/J is a permutation of . Any nonempty lP
component can only decrease the number of CF-PS rules, since LP rules
have no effect on the rules except for ruling out linearizations. The
modified LP component shares this feature with the original one. LP
rules of the new kind can also only decrease the number of possible rules.
The constrained generative power should allow for efficient comput-er
implementations of the framework, which in turn can serve as useful
tools for the design and testing of particular grammars. However, the
theory of GPSG, as described by its creators and as outlined in this
paper, cannot be used directly for efficient implementations. The num-
ber of rules generated by the metagrammar is just too large. The enor-
mous rule proliferation is caused mainly by the combinatorial explosion
of possible categories through feature instantiation.
The Hewlett-Packard GPSG system (1982) and one implementation
by Henry Thompson, which are both based on a pre-ID/LP version of
GPSG, use metarules as metagrammatical devices, but have feature
instantiation built into the processor. Agreement checks, which cor-
respond to the effect of the metagrammatical feature-instantiation prin-
ciples, are done at parse time. As Berwick and Weinberg (1982) have
pointed out, the context-freeness of a grammar might not accomplish
much when the number of rules explodes. The more components of the
metagrammar that can be built into the processor (or used by it as ad-
ditional rule sets at parse time), the smaller the resulting grammar will
be. The task is to search for parsing algorithms that incorporate the
work of the metagrammar into context-free phrase structure parsing
without completely losing the parsing-time advantages of the latter.
Most phrase structure grammar parsers do feature handling at parse
time.
In the process of rule derivation, however, the LP component has to
apply after feature instantiation. Thus, an incorporation of feature in-
stantiation into the actual parsing process depends on a similar incor-
poration of the LP component.
Recently Shieber (1984) extended the Earley algorithm (1970) to in-
corporate the linearization process without a concomitant loss in parsing
efficiency. 58 The redefinition of the LP component proposed in Chapter
5, Section 2 can be integrated easily and efficiently into Shieber's exten-
sion. Even if the number of computational steps that need to be per-
formed when the modified LP component is applied to the input string
might be higher than before (because of complex LP rules), the increase
in complexity is effected not by the length of the input string, but rather
by the number and size of LP rules.
It can therefore be concluded that the treatment of partially free
word order in the slightly modified framework of GPSG has no negative
effects on generative capacity and implementability.
In the long run, the modified LP component will probably support
considerable gains in efficiency. These gains result from the encoding of
stylistic principles as it is required for high-quality language generation.
No additional component is necessary to accommodate stylistic choice.
As was outlined in Chapter 5, Section 3.1, stylistic principles can be
easily implemented by merely specifying the weight of competing atomic
LP rules dynamically with respect to the text or discourse type, then
using these weights in selecting the proper linearization.
3 Expandability
3.1 Additional Verb Subcategorization Frames
The fragment includes intransitive, tr ansitrve, and ditransitive verbs,
along with verbs that take only a dative object. The majority of Ger-
man verbs belong to one of these four classes. Yet the three sub-
categorization frames represent just a small subset of the verb sub-
categorizations that occur in German. Other sub categorization frames
include genitive objects, prepositional objects, infinitival and sentential
complements, and predicative phrases. There are two potential problems
entailed in extending the fra.gment to the entirety of verb subcategoriza-
tions. One of them concerns the features used to specify unmarked order
within the partially free order among the arguments of the verb, while
the other follows from the existence of subjectless verbs in German. I
shall start with the additional features that might be needed to express
the unmarked order.
58Shieber's ID/LP version of the Earley algorithm has been modified and implemented in
PROLOG by James Kilbury (Kilbury (1984)).
Expandability 1:13
3.1.1. The Specification of the Unmarked Order
It was pointed out in Chapter 2, Section 5 that it is an open question
whether an adequate linguistic theory can do without assuming gram-
matical functions such as subject and object are part of the primitive
concepts. It was mentioned that most versions of GPSG do not make
reference to grammatical function in their grammars, but rather treat
these as derivable concepts. In Chapter 1, I have introduced the prin-
ciples of unmarked order in terms of grammatical functions, since this is
the terminology used in most descriptive grammars. Lenerz's tests are
also described by referring to concepts such as subject, direct object, and
indirect object. In GFG, however, the same regularities are encoded in
terms of syntactic categories. Case marking is a syntactic feature and, as
such, part of a category specification. The encoding makes use of a well-
known fact about German: subject NPs are marked nominative, indirect
objects are marked dative, and direct objects are marked accusative. If
case marking and grammatical functions always coincide in this manner,
there is obviously no need for grammatical functions within the LP com-
ponent of the German grammar.
The fragment is easily extended to genitive objects. The unmarked
order positions them after subjects and accusative objects, which are the
only other arguments with which they cooccur. Prepositional objects do
not cause problems either; their unmarked position is also after the other
arguments.i''' The unmarked order in clauses with two prepositional ob-
jects, such as (312), needs to be determined:
(312) Peter sprach zu den Kindern tiber seine Arbeit.
Peter spoke to the children about his work
Peter talked to the children about his work.
If there exists such an unmarked order, the question arises as to whether
the order is determined by the prepositions or is specific to the verb.
Should it turn out to be specific to the verb, an additional feature will be
needed to encode it in the categorial information associated with the
prepositional phrases.
There is more evidence for the need of additional features. Certain
verbs subcategorize for daB-clauses, the German equivalent of English
that-clauses:
59Matters will get more complicated when adverbial phrases are included, since, as Lenerz
(1977) points out, temporal and locative adverbials precede prepositional objects in the
unmarked order.
(313) Der Doktor weiB, daB der Patient kommt.
The doctor knows that the patient comes
The doctor knows that the patient is coming.
DaB-clauses can also occur in subject position:
(314) DaB der Patient kommt, verwundert den Doktor.
That the patient comes amazes the doctor
It amazes the doctor that the patient is coming.
If the unmarked order can be stated in purely syntactic categorial terms
without reference to grammatical funtion, we should expect that all
daB-clauses behave alike with respect to their unmarked position. This
claim is not easily tested, since the preferred position for any sentential
complement or adjunct is outside the clause core. Although the
extraposition process that moves sentential elements to the right of the
clause core is not syntactically obligatory, strong stylistic preferences
might affect the acceptability of the relevant examples.
Example (315b) appears less marked than (315a):
(315)
a. Darum hat, daB der Patient lebt, der Doktor
Therefore has that the patient lives the doctor
gewuBt.
known
Therefore the doctor knew that the patient was alive.
b. Darum hat der Doktor, daB der Patient lebt,
Therefore has the doctor that the patient lives
gewuBt.
known
Therefore the doctor knew that the patient was alive.
If the daB-clause is the subject, the preferred order is reversed. Sen-
tence (316a) sounds less marked than (31Gb).
(316)
a. Darum hat, daB der Patient lebt,
Therefore has that the patient lives
den Doktor uberrascht.
the doctor surprised
Therefore the doctor was surprised that the patient was
alive.
Expandability
13E,
b. Darum hat den Doktor, daB der Patient
Therefore has the doctor that the patient
lebt, uberrascht.
lives surprised
Therefore the doctor was surprised that the patient was
alive.
Although it appears that the unmarked order is influenced by the ar-
gument position of the daB-clause, examples (315) and (316) could also be
explained by atomic LP rules that let daB-clauses follow nominative NPs
and precede accusative NPs. This prediction is not easily falsified, since
for many speakers of Standard German there are no verbs that take an
accusative complement together with a daB-clause. For other speakers,
however, verbs such as auffordern, bitten, warnen fall into this catego-
ry. For those speakers (317) is fully acceptable:
(317) Ich hatte den Geiger gebeten, daB er aufharen mage.
I had the violinist asked that he stop may
I had asked the violinist whether he might stop.
Then the following permutations can be tested:
(318)
a. Ich hatte den Geiger, daB er aufharen mage,
I had the violinist that he stop may
gebeten.
asked
b. Ich hatte, daB er aufharen mage, den Geiger
I had that he stop may the Violinist
gebeten.
asked
The order in (318a) is more natural than the order in (318b). There is
no way to explain this contrast solely in terms of syntactic category and
morphological features. A similar observation can be made for the un-
marked order of daB-clauses and dative NPs. Consider first a sentence
with a clausal subject and a dative object. Again the order that places
the subject before the object sounds more natural:
136
Evaluation and Expendability of the Grammar
(319)
a. Darum hat, daB Peter Reue zeigt,
Therefore has that Peter remorse shows
dem Anwalt geholfen.
the lawyer helped
The fact that Peter showed remorse therefore helped the
lawyer.
b. Darum
Therefore
hat dem Anwalt, daB Peter Reue
has the lawyer that Peter remorse
zeigt, geholfen.
shows helped
The fact that Peter showed remorse therefore helped the
lawyer.
And now let me finally contrast the permutations of a clausal
complement and an indirect object:
(320)
a. Der Doktor hat, daB der Patient lebt,
The doctor has that the patient lives
dem Richter mitgeteilt.
the jUdge told
The doctor informed the judge that the patient was alive.
b. Der Doktor hat dem Richter, daB der Patient
The doctor has the jUdge that the patient
lebt, mitgeteilt.
lives told
The doctor informed the judge that the patient was alive.
Here it seems that the order in which the dative NP precedes the
daB-clause is less highly marked. Again we find an ordering contradic-
tion; sometimes the daB-clause precedes and sometimes it follows a dative
NP in the unmarked order. If every phrase were marked for its gram-
matical function, the unmarked order constraints could be easily stated
with the set of atomic LP rules abbreviated in (321):
(321) SUBJ < IDBJ < DOBJ < daB-COMP
A type of verb complementation that does not allow for the state-
ment of the unmarked order of the complements solely in terms of cate-
gory and case is found in double-accusative verbs. The two objects of
double-accusative verbs clearly exhibit a preferred order:
Expandability 137
(322)
a. Peter hatte die Kinder das neue Gedicht abgefragt.
Peter had the children the new poem ? asked
Peter had the children recite the new poem.
b. ?Peter hatte das neue Gedicht die Kinder abgefr~~.
Peter had the new poem the children ? asked
An additional feature would be needed to state this order. However,
without further research it cannot be determined whether the order of
the two accusative objects can be classified as unmarked order. Viola-
tions of the order cannot be easily legitimized by some competing order-
ing principle; (322b) does not apear fully acceptable even in a context
that makes die Kinder the focus of the sentence. Yet, I think that the
following sentence passes the test:
(323) Peter hatte es die Kinder abgefragt.
Peter had it the children? asked
Peter had the children recite it.
Again, it is obvious that category hood and case do not suffice to
state the unmarked order. The required atomic LP rule would be easily
formulated if the two accusative noun phrases were marked for different
grammatical functions. It needs to be stressed, though, that the ad-
ditional verb complementation types do not provide a conclusive ar-
gument for the inclusion of grammatical functions into the set of syntac-
tic primitives. Nor does the evidence conflict with the framework or
analysis instantiated in GFG. Other linguistically motivated sets of fea-
tures that marked complements according to types of verb arguments
might do as well.
Let me mention another potential argument against the statement of
unmarked order in categorial terms. Lenerz (1977) observes that the un-
marked order of subject and dative-object is reversed in passive
sentences. He cites the following test as evidence.I"
(324)
a. Wem ist das Fahrrad geschenkt worden?
Whom is the bike given been?
To whom has the bike been given?
b. Ich glaube, daB das Fahrrad dem Kind
I think that the bike the child
70Lenerz (1977), p.1l6 (Added glosses, different numbering).
geschenkt worden ist.
given been is.
I think that the bike has been given to the child.
c. Ich glaube, daB dem Kind das Fahrrad
I think that the child the bike
geschenkt worden ist.
given been is.
(325)
a. Was ist dem Kind geschenkt worden?
What is the child given been?
What has been given to the child?
b. Ich glaube, daB dem Kind das Fahrrad geschenkt
I think that the child the bike given
worden ist.
been is.
c. Ich glaube, daB das Fahrrad dem Kind geschenkt
I think that the bike the child given
worden ist.
been is.
This test-together with related observations-leads Lenerz to con-
jecture that the unmarked order of object and subject in passive sen-
tences is identical to the unmarked order of the corresponding con-
stituents in an active sentence, i.e., indirect object and direct object.
Thus, the unmarked order in passive sentences adds to the evidence
against the assumption that the LP rules that govern the unmarked or-
der can be formulated in terms of category hood and case.
The unmarked order in passive sentences could be expressed in terms
of the grammatical functions of the corresponding active sentence.
Whatever type of rule links active and passive sentences in GPSG
71
could
be written in a way that assigns to the subject of passive sentences the
grammatical function of the object. Obviously, such a utilization of
grammatical function would differ radically from the concept of gram-
matical function as it is traditionally used to express such generalizations
as the position of the subject with relative to the verb in fixed order
languages. A framework with several levels of syntactic structure, such
as traditional Transformational Grammar or Relational Grammar, could
mark a constituent subject on one level and an object on another. The
one-level framework of GPSG does not offer this possiblity.
71Th is applies to a metarule approach with and without phantom categories as well as to a
lexical passive rule.
Expandability
The data I want to consider next are derived from verbs with non-
agentive subjects and from passive verbs; they might prove helpful in
selecting the appropriate class of additional features. German-just like
English-has a variety of verbs in which the subject does not represent
the agent. A subclass of these verbs combines with agentive objects.
(326) Das Buch hat dem Mann gefallen.
The book has the man pleased
The man liked the book.
(327) Die Operation ist dem Arzt gelungen.
The operation is the doctor succeeded.
The operation was a success for the doctor.
I do not want to discuss the numerous attempts to unify the properties of
these verbs with the generalizations found with other verbs_72 Lenerz
(1977) observes that the unmarked order for the arguments of agentive-
object verbs often differs from the unmarked order for the corresponding
arguments of other verbs. Consider the following pair of sentences:
(328)
a. Darum war die Operation dem Doktor gelungen.
Therefore was the operation the doctor succeeded
Therefore the operation was a success for the doctor.
b. Darum war dem Doktor die Operation gelungen.
Therefore was the doctor the operation succeeded
Therefore the doctor was successful in the operation.
Example (328b) represents the less highly marked order of subject and
dative object. There are other verbs like gefallen that seem to lie some-
where between verbs that induce the regular unmarked order of SUBJ-
OBJ and verbs that induce the opposite unmarked order. I have tried to
test the unmarked order of the arguments of the verb gefallen which is
usually considered to be in the relevant class of agentive-object verbs. As
for previous tests I have used Lenerz's method.
(329)
a. Wem hatte das Haus
Whomhad the house
Who liked the house?
gefallen?
pleased
72Examples in the recent history of linguistics are Lakeoff's FLIP transformation (Lakoff
(1965)), Postal's transformation of Psych Movement (Postal (1971)), or Perlmutter's
Unaccusative Hypothesis (Perlmutter (1983)). The classes of verbs that are considered by
the different approaches do not overlap fully.
b. Ich glaube, daB das Haus einem Doktor gefallen hat.
I believe that the house a doctor pleased has
I think that a doctor liked the house.
c. Ich glaube, daB einem Doktor das Haus gefallen hat.
I believe that a doctor the house pleased has
I think that a doctor liked the house.
(330)
a. Was hatte dem Doktor gefallen?
What had the doctor pleased
What had pleased the doctor?
b. Ich glaube, daB dem Doktor ein Haus gefallen hat.
I believe that the doctor a house pleased has
I think that the doctor was pleased by a house.
c. Ich glaube, daB ein Haus dem Doktor gefallen hat.
I believe that a house the doctor pleased has
I think it was a house that pleased the doctor.
For the few informants I asked, including myself, there was no detectable
contrast in acceptablility between (329c) and (330c).
Further testing-involving a wider set of verbs and a larger number
of informants-would be necessary to establish clear, verifiable results.
I shall refrain from further speculation about the exact nature of the
concepts that are needed to describe the unmarked order of verb ar-
guments and rather summarize the previous findings. Obviously, case
and categoryhood do not suffice to determine the unmarked order. In a
framework that encodes grammatical functions in the verb entries (as op-
posed to deriving them from the "surface" phrase structure), the reversal
of the unmarked order in passive sentences could be explained if un-
marked order were defined in terms of grammatical functions. The pas-
sive rule simply needs to preserve the assignment of grammatical
functions to the argument positions of the verb. As I pointed out earlier,
the cost would be a radical redefinition of the meaning of grammatical
function.
If one attempted to use grammatical functions to explain the un-
marked order of the arguments of psychological verbs, the use of con-
cepts such as subject and object would be even farther removed from
their original meaning. The sub categorization frames for these verbs (in
traditional versions of GPSG encoded directly in PS rules) would consis-
tently have to mark the subject object and the object subject. There are
no homonymous or morphologically related forms from which the
psychological verbs could be derived by relation-changing (and
grammatical-function-preserving) lexical or syntactic rules.
Let me repeat now in summary the assembled apparent evidence
against relying on category hood and case for encoding the unmarked or-
Expandability 141
der. In so doing, I will also again address the question of whether the
unmarked order should be expressed in terms of grammatical functions.
Clausal arguments of the verb, especially daB-clauses, cannot be or-
dered properly without introducing additional concepts into the relevant
LP rules. Grammatical functions could be employed to encode the order-
ing principles. The unmarked order of subject and dative object in pas-
sive sentences also requires the introduction of different concepts into LP
rules. Grammatical functions could be used for marking the arguments.
However, in such an approach the terms subject and object would mark
argument slots of the verb and would not correspond to the surface sub-
ject and object of the passive sentence. A third argument, against the LP
rules of GFG can be based on verbs with a nonagentive subject and an
agentive object. In sentences with verbs of this kind, the unmarked or-
der places the object before the subject. An attempt to explain this or-
der in terms of grammatical functions within the framework of GPSG
has led to lexical entries for the relevant verbs that mark the subject of
the sentence as the object and vice versa. This is an unintuitive and
theoretically unsatisfactory solution.
The examples show the need of additional features that can be used
by the LP component to impose the unmarked order. Very little
research has been done in this area. An investigation of the complex
problem is beyond the scope of this book. Lenerz (1977) proposes the
notion of Mitteilungszentrum (center of information). His characteriza-
tion of the concept makes it overlap with the meaning that is assigned to
the term theme by contemporary members of the Prague School.
However, most if not all of Lenerz's examples could easily be accounted
for by means of thematic roles. In the cases of passive sentences and of
nonagentive-subject verbs, it is the agenthood that determines the devia-
tion from the regular unmarked order.
A formulation of unmarked-order LP rules in terms of thematic roles
would also allow for an explanation of a difference between the
nonagentive-subject verbs gelingen and gefallen. Both verbs have a da-
tive object, but only gelingen assigns an agent role to the referent of the
object NP. The unmarked order of subject and object for gelingen is the
converse of the regular unmarked order; i.e., the object precedes the sub-
ject. Cefallen differs from verbs with agentive subjects in that there
does not seem to be a specified unmarked order of subject and object.
Without much discussion I will use a single example to indicate that
the agentiveness also plays a role for the unmarked order of daB-clauses
and objects. The following example was chosen earlier (page 136) to
demonstrate that a subject daB-clause can precede a dative object in the
unmarked order:
(331) Darum hat, daB Peter Reue zeigt,
Therefore has that Peter remorse shows
dem Anwalt geholfen.
the lawyer helped
The fact that Peter showed remorse therefore helped the lawyer.
If the dative object is agentive the unmarked order is reversed: example
(332a) is less highly marked than example (332b):73
(332)
a. Darum war dem Arzt, daB der Patient uberleben
Therefore was the doctor that the patient survive
konnte, doch noch gelungen.
could yet still managed
Therefore the doctor managed after all that the patient
would survive.
b. Darum war, daB der Patient uberleben konnte,
Therefore was that the patient survive could
dem Arzt doch noch gelungen.
the doctor yet still managed
In the previous paragraphs I have stated and motivated the assump-
tion that additional features are needed to encode the unmarked order
among arguments of the verb, and that these features should not be
grammatical functions but rather thematic roles. At this point I need to
stress that I do not consider obsolete the atomic LP rules in GFG that
encode the unmarked order in terms of grammatical case. I suggest that
these LP rules interact with atomic LP rules that are based on thematic
roles, e.g., a rule that lets an agent NP precede a patient NP. Such an
interaction would account for the unmarked order in sentences with so-
called symmetric verbs such as begegnen (meetf4 and ahneln (resemble).
Although subject and object of a symmetric verb fill the same thematic
role, the unmarked order, as Lenerz observes, positions the subject before
the object.
Lenerz also notes that so-called free datives, dative NPs that are free
adjuncts, behave like dative objects with respect to their place in the un-
marked order. One subclass of these free datives contains NPs with a
benefactive thematic role.
73See page 134 for a remark on the low acceptability of sentences with nonextraposed
daB-clauses.
74This is not to say that the English verb meet is symmetric; actually it is not.
Expandability
143
(333) Peter hatte einem Freund einen Koffer getragen.
Peter had a friend a suitcase carried
Peter had carried a suitcase for a friend.
The free dative preceding the object is the unmarked order. The
benefactive NP could be replaced by a benefactive PP. Lenerz
(1977) claims that the behavior of the PP in the unmarked order
resembles the behavior of prepositional objects. Whereas the free dative
tends to precede the object, the corresponding prepositional phrase shows
a tendency to follow it. An interaction of atomic LP rules that are based
on case and categoryhood with atomic LP rules that mention thematic
roles could account for the unmarked position of free datives.
Finally I can summarize the foregoing discussion. There were ap-
parent reasons to assume that an extension of GFG to wider classes of
verb complementations would not be compatible with the GFG approach
to unmarked order. It seemed, moreover, that some of the criti.cal
phenomena could be problematic for the theory of GPSG, since gram-
matical functions appeared necessary for encoding unmarked order.
Without reformulating or extending the grammar, I showed that the
critical constructions require merely the addition of LP rules based on
thematic roles or similar concepts. I did not propose a specific set of
features nor did I suggest how these features should be introduced.
3.1.2. Subjectless Verbs
A small class of German verbs occurs without a subject, i.e., without a
nominative noun phrase. All subjectless verbs are sensation verbs. The
experiencer of the sensation is represented by a dative or accusative
noun-phrase.
(334)
a. Dem Patienten graute vor der Medizin.
The patient horrified before the medicine
The patient was horrified by the medicine.
b. Ihn fror.
Him froze
He was freezing.
c. Den Gefangenen durstete.
The prisoner thirst had
The prisoner was thirsty.
The existence of subjectless verbs could be taken as evidence for sug-
gesting that the basic rules introducing verbs should be clause rules.
This approach would imply the statement that German verbs subeat-
gorize for subjects. If subjectless verbs were introduced by VP rules, a
144
special feature would be needed that distinguishes their VP rules from
other VP rules that combine a verb with a dative or accusative noun
phrase. The metarule that derives clause rules from verb phrase rules
had to be made sensitive to this feature. In traditional transformational
grammar, features of this kind are called exception features. There are
strong reservations against such otherwise unmotivated features. The
same is true for GPSG.
But the required exception feature is not the only disadvantage of
VP rules for subjectless verbs. Although the clauses under (334) have the
same constituents as ordinary VPs, they cannot function as VPs.
(335)
a. Der den Jungen stOrende Wind wurde starker.
the the boy bothering wind became stronger
The wind that was bothering the boy became stronger.
b. *Der den Jungen frierende Wind wurde starker.
the the boy chilling wind became stronger
(336)
a. Der Doktor versuchte, dem Patienten zu helfen.
the doctor tried the patient to help
The doctor tried to help the patient.
b. *Das Gespenst versuchte, dem Kind zu grauen.
the ghost tried the child to terrify
However, if all the basic rules that introduce verbs were clause rules,
the right set of VP rules could be derived by a metarule:
(337) V3 --+ N
2
,
+NOM
x ==} V
2
--+ X
Metarule (337) has to presuppose that predicative nominative noun
phrases are dominated by a different category. There will be no verb
phrase rules introducing subjectless verbs.
Although this option might have some advantages that I am not yet
able to perceive, I will defend the GFG approach for the time being. Ex-
cept for the existence of a handful of subjectless verbs, all arguments
against subcategorizing for subjects in English carryover to German.
Therefore, I simply suggest to introduce subjectless verbs by basic clause
rules directly.
(338) (35; V3 --+ N
2
,V) dursten, frieren, gelusten
+ACC
Expandability 145
(339) (36; y3 ~ N
2
,V) grauen, gruseln, schaudern
+DAT
Highly idiosyncratic phenomena require highly idiosyncratic solutions. In
GPSG, the idiosyncracies in verb subcategorization are reflected in basic
rules that correspond to the assumed subcategorization frame. The num-
ber of double-accusative verbs in German is no larger than the number of
subjectless verbs, yet there needs to be a special rule that combines a
verb with two accusative objects.
There is no evidence that subjectless verbs ever occur in verb
phrases. The assumption of basic clause rules for subjectless verbs seems
sufficiently justified.
3.2 Adverbial Phrases
In Uszkoreit (1982a) certain potential complications were mentioned Chat
followed from the integration of adverbial phrases into the flat-structure
approach. The problem stems from the fact that adverbial phrases can
be interspersed freely among the arguments of the verb.
(340) Gestern hatte in der Mittagspause der Vorarbeiter
yesterday had during lunch break the foreman
in der Werkzeugkammer dem Lehrling aus Boshaftigkeit
in the tool shop the apprentice maliciously
langsam zehn schmierige Gusseisenscheiben unbemerkt
slowly ten greasy cast iron disks unnoticed
in die Hosentasche gesteckt.
in the pocket put.
Yesterday, during lunch break, the foreman maliciously and
unnoticed put ten greasy cast iron disks slowly into the
apprentice's pocket.
Stucky (1981) solves the similar problem for Makua by formulating a
metarule that makes the adverbial phrase a sister node of the verb and
its arguments. This would be the corresponding metarule for GFG:
(341) V
2
~ X ==} V
2
~ X, AdvP
-AUX
For every rule that introduces a main verb into a verb phrase, there
is also a rule just like the input rule, except for an additional category
AdvP on the right-hand-side. Since there can be more than one adver-
bial phrase inside the verb phrase (or clause), the metarule has to apply
recursively. Recursive applications of metarules are ruled out, however,
by the constraint of Finite Closure (see Chapter 2, Section 6). The con-
straint was introduced to prevent the formalism from generating infinite
rule sets, and, moreover, super-CF object grammars. The distribution of
adverbial phrases does not induce a non-CF language, since only one cat-
egory is added recursively and the metarule does nothing else besides ad-
ding this category.
But let us imagine a grammar with one basic rule (342) and one
metarule (343), which adds several symbols to the rule every time it ap-
plies:
(342) A -4 a, b, C
(343) A -4 X =} A -4 X, a, b, C a, b, C E V
T
In addition, imagine LP rules that encode the following partial order on
the vocabulary:
a<b<c
Then the resulting language is an b
n
c", a well-known non-CF language.
Some versions of GPSG make use of the Kleene star notation, as it is
used in regular expressions. Gazdar et al. (1982b) employ the notation 75
to create flat structures for coordinate conjunction. The violation of
Finite Closure can be circumvented by adding the following metarule:
(344) V
2
-4 X =} V
2
-4 X, AdvP*
-AUX
As long as the scope of the Kleene star is restricted to a single category
symbol, the notation will not give rise to non-CF languages (although the
object grammar will contain an infinite rule set). Otherwise it will have
the same effect as the fictitious metarule (343).
Lenerz (1977) and Heidolph et al. (1981) agree with previous descrip-
tive grammars about the existence of an unmarked order among free ad-
verbial phrases as well as among the latter and arguments of the verb.
The unmarked position of an adverbial phrase is detemined by its class.
The classes to a considerable degree correspond with the traditional clas-
sification of adverbial phrases into temporal, local, manner, and other
types. It is a consequence of the decision to treat free adverbials as si-
blings of the verb and its arguments that requires encoding of the clas-
sification in the complex categorial symbol. Their order is not deter-
75 Actually, the conjunction rule schema uses +, the positive closure, instead of *.
Expandability
1-17
mined by semantic restrtctions on their order of application to a verb
phrase or a clause, but rather by LP rules that order sibling symbols in
phrase structure rules.
3.3 Expletive es
In Chapter 1, Section 5 and in Chapter 3, Section 4, the constituents
were characterized that occupy the first position in verb-second clauses.
In assertion main clauses, the clause initial position is filled by a topical-
ized phrase. The topicalized element is either the topic of the sentence or
it is a contrastive or emphatic focus. However, the clause initial con-
stituent may also be the expletive pronoun es. The form of es I am
referring to needs to be distinguished from the regular personal pronoun,
as in (345).
(345) Es schlaft.
It sleeps
It is asleep.
Here es is the subject. It refers to someone or something. Another
form of es can be found as the subject of weather verbs or other verbs
that can denote agentless events.
(346) Es regnet.
It rains
It is raining.
(347) Es trug ihn aus der Kurve.
It carried him out the curve
He was thrown off the curve.
A third use of es makes it a resumptive pronoun for extraposed
clauses.
(348) Es ist gut, daB Peter kommen konnte.
It is good that Peter come could
It is good that Peter could come.
The three uses of es that are exemplified in (345) through (348) are not
restricted to clause-initial position:
(349) Darum schlaft es.
Therefore sleeps it
Therefore it is asleep.
148
(350) Darum regnet es.
Therefore rains it
Therefore it is raining.
(351) Darum trug es ihn aus der Kurve.
Therefore carried it him out the curve.
Therefore he was thrown off the curve.
(352) Darum ist es gut, daB Peter kommen
Therefore it is good that Peter come
Therefore it is good that Peter could come.
konnte.
could
The clause-initial occurrence of these forms of es is covered by GFG
through the topicalization rule. The case I want to discuss here is the
use of the expletive pronoun to fill the first position in verb-initial
clauses in which no meaningful constituent was fronted. These sentences
may have a subject:
(353) Es kamen zwei Manner zur TUr herein.
It came two men at door in
Two men came in through the door.
They might also lack a subject, such as so-called impersonal passives?6
(354) Es wurde bis zum Morgen getanzt.
It was till to-the morning danced
There was dancing until morning.
In both cases, the expletive es is confined to clause-initial position.
(355) *Zwei Manner kamen es zur TUr herein.
Two men came it at door in
(356) *Bis zum Morgen wurde es getanzt.
Till to-the morning was it danced
If the first position is filled by a meaningful constituent, the expletive
. b 77
pronoun IS a sent.
76
For
a description and analysis of the interaction between impersonal passive and expletive
es, see Nerbonne (1982b)
77
piitz
(1975) presents an extensive distributional analysis of the pronoun es, The result is
an exhaustive classification of different uses. The class singled out here corresponds to
Piitz's final class 1.
Expandability
149
(357) Zwei Manner kamen zur TUr herein.
Two men came at door in
Two men came in through the door.
(358) Bis zum Morgen wurde getanzt.
Till to-the morning was danced
There was dancing until morning.
There is no reason to believe that es is an argument of the verb. The
expletive pronoun has no independent meaning. It would be the only ar-
gument whose distribution is restricted to the initial position of assertion
main clauses. It is inconceivable that there should be any verb ar-
guments that cannot occur in subordina.te clauses. A similar argument
applies to a treatment of the expletive es as an adverbial phrase.
There is an easy solution. If the role of the expletive pronoun is so
special, it should receive special treatment in the syntax. The following
rule could be added:
(359) -y3 -+ es -y3
+AC +MC
+FIN -AC
Rule (359) introduces es syncategorematically by combining it wit.a a
main clause that, by itself, is not an assertion clause, i.e., with a verb-
initial clause. An LP rule is needed that ensures that the pronoun will
precede the clause.
Rule (359) generates the expletive pronoun only in the right places.
The existence of such a rule could be explained as follows. Speakers
want to use sentences as assertion clauses in which no phrase is topical-
ized. The language provides a way to make these clauses into assertion
clauses by giving them the verb-second pattern without any extraction.
However, this explanation does not answer the question of whether any
phrase can be combined with the expletive pronoun. Without Rule (359),
topicalization was forced. Does the new rule make topicalization op-
tional? The rule might only be invoked if no phrase can be topicali zed.
Or it might be restricted to clauses in which no phrase needs to be topi-
calized. Unfortunately, not enough is known yet about Germa
topicalization to answer these questions. However, there are clear indica-
tions that Rule (359) will not suffice without further modifications.
Consider the following examples:
(360)
a. Es kam ein Mann zur TUr herein.
It came a man at door in
A man came in through the door.
150
b. ?Es kam der Mann zur TUr herein.
it came the man at door in
The man came in through the door.
c. Es kam der gleiche Mann zur TUr herein
it came the same man at door in
The same man came in through the door.
d. *Es kam er zur TUr herein.
it came he at door in
The expletive es does not combine with clauses that have personal
pronouns as subjects. This observation is not restricted to agentive
subjects:
(361)
a. Es fehlte den Kindem der Vater.
it missed the children the father
The children missed their father.
b. Darum fehlte er den Kindern.
Therefore missed he the children
Therefore the children missed him.
c. *Es fehlte er den Kindem.
It missed he the children
And it does not extend to agentive von-phrases (by-phrases) III passive
constructions:
(362) Es wurden von ihm Viele Kinder getauft.
It became by him many children baptized
Many children were baptized by him.
On the other hand, consider the following expletive pronoun clauses with
pronominal object:
(363) Es besuchten ihn viele Leute.
+ACC
It visited him many people
Many people visited him.
(364) Es halfen ihnen aIle Nachbarn.
+DAT
It helped them all neighbors
All the neighbors helped them.
It was noted earlier that further research is needed to determine the
relationship between pronominalization and discourse roles. Purely syn-
Expandability
151
tactic approaches are bound to be guesswork. Too little is known about
the interdependencies of syntactic properties of constituents and their
discourse functions to permit a closer investigation of the expletive es
within the scope of this book.
3.4 Focus Raising
Uszkoreit (1982b) presents evidence for a class of long-distance filler-gap
dependencies in German that cannot be subsumed under constituent
question formation or topicalization as it was described in Chapter 1,
Section 5 and in Chapter 3, Section 4. In these constructions, the frller
constituent does not occupy clause-initial position. In each of the follow-
ing examples, a constituent has been extracted and "moved" up the tree.
(365) Letztes Jahr hatte Peter das groBe Haus der Stadt
Last year had Peter the big house the city
versprochen zu reparieren.
promised to repair
Last year Peter had promised the city to repair the big hOUSE.
(366) Darum hatte ich dieses Fahrrad den Kindem
Therefore had I this bicycle the children
versprochen bis morgen zu reparieren.
promised until tomorrow to repair
Therefore I had promised the children to try to repair this
bicycle by tomorrow.
Extractions of this type can cooccur with ordinary Wh-Movement depen-
dencies.
(367) Aus welchem Grund hast du das groBe Haus der Stadt
For what reason has you the big house the city
versprochen zu renovieren?
promised to renovate
For what reason did you promise the city to renovate the big
house?
Speakers prefer nested dependencies over crossed ones.
152
(368)
a. In diesen Kasten
1
hatte er dem gleichen Affen
2
In this box had he the same monkey
versucht --2 beizubringen, --1 seine
tried to teach his
Bananenschalen zu werfen.
banana peels to throw
He had tried to teach the same monkey to throw his
banana peels into this box.
b. ??Dem gleichen Affen
1
hatte er in diesen Kasten
2
The same monkey had he in this box
versucht --1 beizbringen, --2 seine
tried to teach his
Bananenschalen zu werfen.
banana peels to throw
Uszkoreit (1982b) calls the extraction process secondary
topicalization. As was pointed out to me by several colleagues, the term
topicalization is quite misleading. It is probably even less appropriate in
this context than it is for ordinary topicalization. Whereas the first posi-
tion of a sentence might actually be filled by a topic constituent, the
fillers of the second type of dependencies are always a focused con-
stituent. It is probably even restricted to emphatic and contrastive focus
constituents. For the time being, I shall refer to the process as focus
raisinq.
The filler-gap relation is a long-distance dependency. In (369) the
extracted phrase is "moved up" over two VP nodes:
(369) Ich kann dir das Rad nicht geben, weil ich dieses
I can you that bike not give because I this
Fahrrad den Kindem versprochen habe zu versuchen
bike the children promised have to try
bis morgen zu reparieren.
until tomorrow to repair
I cannot give you the bike because I promised the children to try
to repair this bicycle by tomorrow.
In Section 3.5, it was mentioned that Wh-Movement type processes in
German have been considered clause-bounded. Counterexamples were
Expandability 11>3
cited. The constraint of clause boundedness applies even more strictly to
focus raising. It is an interesting fact about the latter that the position
of the extracted constituent is influenced by the same LP rules that
determine the order of the surrounding constituents of the host clause.
Consider, again, Example (366), repeated here as (370):
(370) Darum hatte ich dieses Fahrrad den Kindem
Therefore had I this bicycle the children
Therefore I had promised the children to repair this bicycle by
tomorrow.
A less marked variation is (371) in which the unmarked order +DATIVE
< +ACCUSATIVE is observed:
(371) Darum hatte ich den Kindem dieses Fahrrad
Therefore had I the children this bicycle
Therefore I had promised the children to repair this bicycle by
tomorrow.
As in English and German Wh-Movement, the process of focus raising
can extract certain postnominal prepositional phrases:
(372) Gestem haben wir von diesen Autoren wieder viele
Yesterday have we of these authors again many
BUcher verkauft.
books sold
Yesterday we again sold many books by these authors.
The next examples demonstrate a difference between focus raising and
German topicalization.
(373)
a. *Dann hatte ich den Kindem zu re:earieren
Then had I the children to repair
versprochen das Fahrrad bis
morgen _____ .
promised the bicycle until tomorrow
b. *Dann hatte ich zu re:earierenden Kindem
Then had I to repair the children
versprochen bis morgen das Fahrrad
promised until tomorrow the bicycle
Focus raising is less acceptable if the case marking of the filler is the
same as the case marking of some major constituent of its most clause:
(374) ?Ich hatte ihm darum diesen Kindem versprochen
I had him therefore these children promised
zu helfen.
to help
I had therefore promised him to help these children.
(375) ?Ich hatte sie deshalb dieses Buch gebeten bis
I had her therefore this book asked until
morgen zurUckzubringen.
tomorrow to retum
I had therefore asked her/them to return this book by
tomorrow.
Instead of going into further detail with regard to the interesting
phenomenon of focus raising, I shall just try to show that GFG can be
extended to accommodate the observed interaction between a long-
distance filler-gap dependency and the principles that govern the par-
tially free constituent order. The basic solution is extremely simple. A
slash introduction metarule maps sentence rules into other sentence rules
that generate sentences with filler constituents.
(376) S/j3 -+ X =} S -+ j3 X
-AC
+MC
The feature -AC ensures that the clause will not be an assertion main
clause, i.e., that the clause will not immediately dominate a topic and
another S. The feature +MC excludes rules that combine a sentence with
an auxiliary or modal verb. The slash mechanism is used to link filler
and gap. However, there are two reasons that speak against using the
slash feature to encode a missing focus-raised constituent. One is the
cooccurrence of topicalization and focus raising as it is exhibited in (367).
Under the current definition, only a simple category can be the value of
the feature slash. The second reason follows from the different con-
Expandabili ty 155
ditions on the selection of extracted categories. One difference was men-
. d l' 1 . li . b 78
none ear ier: on y topica ization can II move II ver s.
The first objection cannot be maintained, for focus raising belongs to
the class of multiple extraction phenomena. In the following sentence
two arguments of zuruckoerkau jen are raised:
(377) Dann hatte er den Bestohlenen die gleichen BUcher
then had he the theft victims the same books
versucht zu Schleuderpreisen zuruckzuverkaufen.
tried to dumping prices back-to-sell
Then he had tried to sell the same books to the theft victims
again at dumping prices.
This means that, even if a different category-valued feature for focus
raising is created, this new feature needs to allow for lists, sets, or bag/
9
of categories as its values. Thus, the formalism needs to be extenced
anyway to contain features of this kind. The second objection remains
valid. However, it has to be confronted with a reason for merging the
features of the different types of long-distance dependencies. The
evidence comes from the clear preference for a nesting of topicalization
and focus raising as it was exemplified in (368).
Proper nesting of dependencies can be achieved by storing the ex-
tracted categories in a stack. A stack is equivalent to a list if elements
can only be attached to and removed from one side of the list.
8o
The
question of the right number of long-distance dependency features
remains unanswered. I am currently not in a position to justify one al-
ternative over the other. Further research on the topic will certainly
have to consider similarly unsolved problems in other languages with
multiple extractions with and without nesting constraints or nesting
preferences. .
But let me just assume for a moment that a nesting constraint
should be part of the grammar and that the framework allows for stack-
valued features. Then the same feature could be used for topicalization,
wh-fronting, and focus raising. The differences among the three
processes as to the set of categories that can be extracted could be en-
coded in the so called slash-introduction metarules, i.e., metarules that
combine the filler with the constituent that carries the gap. Only one
78This fact about topicalization distinguishes it also from wh-fronting.
79Bags, or multisets, are sets that allow for multiple memberships of elements.
80Stack features have been discussed in connection with nesting preferences in Scandinavian
unbounded dependencies. See Maling and Zaenen (1980).
slash-termination metarule would be needed. A solution of this kind
could therefore at the same time account for the differences between
topicalization and wh-movement without more than one slash feature be-
ing assumed.
3.5 Additional Types of Topicalization
The analysis of verb-second clauses that was presented in Chapter 3, Sec-
tion 4, allows for the fronting of a single major constituent. The follow-
ing permutational variants can be generated:
(378)
a. Peter wollte dem Jungen das Buch schenken.
Peter wanted the boy the book give
b. Dem Jungen wollte Peter das Buch schenken.
the boy wanted Peter the book give
c. Das Buch wollte Peter dem Jungen schenken.
the book wanted Peter the boy give
d. Schenken wollte Peter dem Jungen das Buch.
give wanted Peter the boy the book
Peter wanted to give the book to the boy.
However, for most speakers of German the following variant is also ac-
ceptable:
(379) Das Buch schenken wollte Peter dem Jungen
the book give wanted Peter the boy
Here the main verb and the direct object are fronted. It is even possible
to front both objects and the main verb:
(380) Dem Jungen das Buch schenken wollte Peter
the boy the book give wanted Peter
In GFG, the fronted strings in (379) and (380) are not analyzed as
constituents. They can therefore not be fronted. To find a solution, one
of the following two strategies must be pursued: either the fronting
mechanism is extended to allow for the extraction of more than one con-
stituent or an analysis is arrived at that generates all fronted strings as
constituents. I do not know of any conclusive investigation of this
problem. Arguments for multiple-constituent fronting could be elicited
by looking at other apparent examples of such a process. One type of
fronting involves a duplication of a fronted constituent by a demonstra-
tive proform.
Expandability ].57
(381)
a. Dem Jungen, dem haben wir geholfen.
the boy him have we helped
We have helped the boy!
b. In der Kiste, da habe ich noch diimmere
in the box there have I still sillier
Beispielsatze.
examples
In the box, I have even sillier sample sentences.
There are also sentences in which the fronted constituent seems to be
preceded by a particle that does not modify the entire sentence.
(382)
a. Aber nur zwei Kinder konnten in die Kiste schauen.
but only two children could in the box look
However, only two children could look into the box.
b. Selbst die Eltern kamen nicht zum Ziel.
even the parents came not to-the goal
Even the parents did not reach the goal.
They have counterparts in English.
(383) Even his shoes he could not find.
Another apparent violation of the verb-second constraint involves the
fronting of two prepositional phrases.
(384)
a. Letztes Jahr in Heidelberg hatte er neue Freunde
last year in Heidelberg had he new friends
gefunden.
found
He had found new friends last year in Heidelberg.
b. Yom Parkplatz tiberdie Brtickezum Haupteingang
from parking lot over the bridge to main entrance
lauft man keine 10 Minuten.
runs one no 10 minutes
To walk from the parking lot, over the bridge to the main
entrance, does not quite take 10 minutes.
Wunderlich (1983b) argues convincingly for an analysis in which these se-
quences of adverbial phrases combine to form complex adverbial phrases.
Any attempt to extend fronting to multiple constituents will have to
face the serious question how multiple-constituent fronting can be con-
strained to the few cases that were discussed here. Sentences like (385a)
to (385c) have to be ruled out.
(385)
a. *Peter dem Jungen wollte das Buch schenken.
Peter the boy wanted the book give
b. *Dem Jungen das Buch wollte Peter schenken.
the boy the book wanted Peter give
c. *Peter das Buch wollte dem Jungen schenken. ,
Peter the book wanted the boy give
The known facts suggest that one should follow the second research
strategy, i.e., design grammars in which every fronted string is analyzed
as a single constituent. In the case of (379), this means that the main
verb and the direct object form a constituent. The fronting in (380) sug-
gests the existence of a verb phrase node in the sentence. However, a
branching clause structure does not permit the utilization of the ID/LP
mechanism for the ordering of the verb and its arguments.
An interesting compromise was proposed by Nerbonne (1982a).
Nerbonne's analysis is inspired by the phantom category analysis of the
English passive (1982). In this analysis, the metagrammar contains ID
rules that expand transitive verb phrases (TVP). However, since there
are no rules that mention TVPs on their right-hand sides, the TVP rules
are never used by the object grammar in any successful derivation. This
does not render them useless, though: TVP rules serve as input to
metarules that generate passive and active verb phrase rules. Thus,
phantom categories are categories that are not recognized as constituents
in the language, but that nevertheless play an important part in the
grammatical description.
If a VSO language does not exhibit VP constituents, the category VP
might still be used as a phantom category in a GPSG for the language.
In German, the category VP is not a phantom category. Although Ger-
man clause structure is flat, VPs appear as infinitival complements, ad-
juncts, and prenominal participle groups. Nerbonne's proposal combines
the flat-structure analysis of Uszkoreit (1982a) with an optional branch-
ing structure.
Verb phrase rules and transitive-verb phrase rules can be used in two
ways: they can serve as input to metarules that derive a flat clause struc-
ture; they can also generate VPs or TVPs within an otherwise flat clause
structure. Therefore, neither VPs nor TVPs are phantom categories.
Both can be realized as constituents and both can be fronted by a
slightly revised topicalization rule. This permits the generation of (379)
and (380). I will not go into the technical details of Nerbonne's solution
here, but procede instead to its most severe deficiency immediately.
Expandability
Nerbonne's rules generate precisely those frontings that are described
in Heidolph (1981). There the fronting of the indirect object together
with the main verb is rules out (p. 721):
(386) Seiner Tochter erzaehlen kann er es mit ruhiger Stimme.
his daughter tell can he it with calm voice
If (386) appears ungrammatical, this may be due to the choice of the
noun phrases. Example (387) is perfectly acceptable:
(387) Dem Jungen schenken wollte Peter das Buch.
the boy give wanted Peter the book
Peter wanted to give the book to the boy.
In general, topicalizations of verbs together with indirect objects are
more often rejected than their counterparts involving direct objects.
However, it would be inappropriate to rule out the full class of such
frontings.
81
Another type of apparent multiple-constituent fronting that cannot
be accounted for by Nerbonne's approach involves sequences containing
more than one verb:
(388)
a. Schenken wollen wird Peter dem Jungen das Buch.
give want will Peter the boy the book
Peter will want to give the book to the boy.
b. Zu trinken aufhoren wollte er im neuen Jahr.
to drink quit wanted he in new year
To quit drinking was what he wanted to do in the new year.
c. Angerufen haben konnte er auf jeden Fall.
called have could he in every case
In any case, he could have called.
Fronted sequences of verbs can be accompanied by complements or ad-
juncts:
(389)
a. Der Stadt verkaufen mussen wird der Onkel aIle
the city sell must will the oncle all
81The variation in acceptability of these topicalizations is discussed in Chapter 3, Section
2.2.
seine Hauser.
his houses
The oncle will have to sell all his houses to the city.
b. Nach hause gehen konnen wird er noch lange nicht.
to home go can will he still long not
He won't be able to go home for quite a while yet.
c. Noch schneller gefahren sein muBte er mit dem
still faster gone have must he with the
neuen Wagen.
new car
He should have gone even faster with the new car.
The questions are: what are the syntactic categories of the fronted con-
stituents in (379), (380), (387), (388), and (389), and how are their
denotations computed. Among all current approaches to German that I
am aware of, only Steedman (1983) could generate the full variety of pos-
sible frontings. Steedman's categorial grammar is based on Steedman
and Ades (1982) and Geach (1972).82
A thorough investigation of the whole range of constituent frontings
will also need to consider other well-known phenomena that are problem-
atic for standard assumptions about the constituent structure of verb
phrases: lassen-constructions, auxiliary preposing in subordinate clauses
before double infinitives, and verb-verb compounding such as
sitzenbleiben.
Prior to such an investigation, no predictions can be made as to
whether GFG could be extended to cover these nonstandard topicaliza-
tions. It should be noted that there is a variety of possible effects on the
grammar. It might be that the clause structure is not as flat as GFG
generates it. In this case a number of solutions to other problems had to
be changed. It might also be that nested structures are needed in ad-
dition to flat structures, related to the latter by metarules or other
metagrammatical devices. One possible tool for creating the required
nonstandard constituents could be contraction metarules (see Chapter 2,
Section 7). However, solutions that do not necessitate employment of
this rule type would be clearly preferable. A third possibility is that the
relevant types of topicalization turn out to belong to a class of processes
that cross traditional constituent boundaries, such as gapping and certain
kinds of sentence ellipsis, and therefore have to be treated in a com-
pletely different way.
82Since this chapter was written, I learned about two more recent proposals to solve this
problem: Johnson (1984) and Nerbonne (1984).
Chapter 7
Conclusion
1 A Selective Summary
The research reported in this book has concentrated on the application of
GPSG to the description of central problems in German syntax. The
results fall into a variety of different categories. Some positive answers
were provided to quite fundamental questions about the adequacy of the
framework with respect to a language that lies somewhere between
English and Warlpiri on the scale of ordering freedom. Contributions
were made to the development of the theory. The proposed modification
of the LP component does not only allow for an adequate description of
competing ordering principles as they are found in many languages, it
also provides the basis for a more plausible characterization of regur-
larities across languages with different degrees of ordering freedom. The
extension of the formalism adds desirable descriptive power without ad-
ding to the weak generative capacity, i.e., it does not increase the class of
languages that can be generated.
The adequacy claim for the framework including the proposed
modification is substantiated in a grammar for a fragment of German
that exemplifies the relevant properties of German word order and COIl-
stituent structure. The grammar incorporates analyses for a number of
phenomena in German.
The discussion of particular problems in German contributes relevant
observations to current linguistic debates. There is, for instance, the
question, which sets of syntactic concepts are relevant for the description
of ordering regularities. This question plays a role for determining the
primitive concepts of adequate linguistic theories. The investigation of
the unmarked order and of apparent subject-object asymmetries has
shown that an interaction of thematic roles and categorial information
governs contrasts that have been attributed to grammatical relations
such as subject and object.
161
162
Conclusion
The discussion of separable prefix construction offers evidence for
non-structure-building lexical incorporation rules. The modified LP com-
ponent provides an explanation for the close interaction of grammatical
and stylistic regularities in the choice of ordering variants. Criteria are
provided for distinguishing stylistic principles from grammatical ones.
The grammar for the fragment has a number of desirable properties.
First of all, it is small and simple. Some idiosyncratic properties of Ger-
man such as the verb-second constraint and the separation of verbs and
prefixes do not have to be stated in the grammar at all, since they follow
from other encoded generalizations that are not limited to one language.
The burning question is whether these and other nice properties of
the grammar can be preserved when the coverage of the grammar is ex-
tended to the rest of Standard German. It is unlikely that the the full
answer to the question could be known in the near future. However, a
serious attempt is made in the previous chapter to give an indication of a
partial answer by considering some selected interesting phenomena that
are not part of the fragment. The results give good reason to be optimis-
tic. They also indicate critical problems for future research.
2 Directions for Further Research
This is only a small selection of subjects for future research that seem to
be prerequisites for a fruitful continuation of the project. Much more
work needs to be done in the area of determining the whole variety of
the ordering principles that govern the partially free order of con-
stituents. By far not enough is known about the interaction of
pronominalization, definiteness, and discourse role assignment. The same
is true for the interaction of discourse role assignment and intonation.
Further research on apparent noncategory fronting as it was dis-
cussed in Chapter 6, Section 3.5 is needed in order to determine whether
alternative constituent structures are needed or whether the phenomenon
belongs to the class of processes that defy traditional constituent
boundaries-such as gapping.
The class of Focus Raising constructions needs to be investigated in
much more detail. This investigation includes the search for the exact
conditions that trigger Focus Raising.
An increasingly important topic in German syntax is the impact of
thematic roles for word order, extraction, and relation changing rules.
Recent work by Wunderlich (HJ84) has started a new era in this discus-
sion. At many instances in this book, syntactic regularities were
described that are governed by thematic role assignment.
After this brief selection of topics for further research on linguistic
phenomena, let me finally mention similar topics for future work on the
formalism and theory of GPSG. The most urgent step in the future
development of GPSG is the design of a more sophisticated and more
Directions for Further Research
Ie.3
powerful lexical component. The need for this component becomes ob-
vious in the investigation of separable prefix verbs. The lexical rules
proposed in Chapter 4 need to be integrated. Two other developments
support the need for a strong lexicon. Uszkoreit and Peters (1983) and
Shieber et al. (1983a) cite reasons for abolishing metarules altogether.
Dan Flickinger's Lexical Head Constraint (1983) requires metarules to
apply to rules with lexical heads. This already covers a wide class of
metarules. There is no reason to assume that these rules could not be
reformulated as lexical rules.
In Chapter 2, Section 6 it was pointed out that it is not known
whether Uszkoreit and Peters' results (1983) about formal properties of
metarule grammars carryover to the ID/LP version of the framework.
The answer to this question might render the Finite Closure Constraint
obsolete.
The last direction for future extensions of the framework to be men-
tioned here concerns the semantics. The suggested solution to the inter-
face of syntax and pragmatics in Chapter 5, Section 3 assumes facilities
for encoding the nontruthconditional elements of the semantics. It was
noted that these facilities do not yet exist. Some general options are sug-
gested. The nontruthconditional part of semantics is not only needed for
the encoding of presuppositions that arise through the lexicon or through
word order, it is also required for the interface between syntax, seman-
tics, and pragmatics in the analysis of certain discourse-dependant syn-
tactic constructions such as cleft, pseudo-cleft, left dislocation,
extraposition.
Most of the critical problems raised in this section have been recog-
nized as theoretical challenges for quite a long time. The purpose of the
list was mainly to indicate areas to which the book has contributed ad-
ditional questions to be asked. To me it seems of no great importance
whether this book has raised more questions than it has answered or
whether the opposite is true-as long as both kinds of questions are
equally interesting.
Rules of GFG
A.I Basic Rules
1. (1, V3 ---+ V, V
2
)
+PERF +PSP
+AUX
2. (2, V
2
---+ V, V2)
+AUX +BSE
3. (3, V
2
---+ V, V
2
)
+AUX +PAS
4. (4, V
2
---+ V, V
2
)
+AUX +BSE
+FIN
5. (5, V
2
---+ V)
6. (6, V
2
---+ V, N
2
)
+ACC
7. (7, V
2
---+ V, N
2
)
+DAT
Appendix A..
(haben, sein)
(muss en, konnen, dur fen, ... )
(werden)
(werden)
(kommen, laufen, schlafen)
(come) (run,go) (sleep)
(Kennen, such en, nehmen)
(know) (search) (take)
(helfen, begegnen, vertrauen)
(help) (meet) (trust)
165
166 Rules of GFG
8. (8, V
2
---+ V, N
2
, N
2
)
+ACC +DAT
(geben, schicken, zeigen)
(give)(send) (show)
9. (9, V
2
---+ V, V3)
+daB
(wissen, glauben, bezweifeln)
(know) (believe) (doubt)
10.(10, V3 ---+ daB, V3)
+daB -MC
11 . (n; SEPREF ---+ 0')
+0'
12.(11; V3 ---+ 0', V3/0')
+AC +TOP +MC
+FIN
A.2 Metarules
A. V
2
---+ X =} V3 ---+ N
2
X
+NOM
B. V
2
---+ V, V
2
=} V3 ---+ V, V3
+AUX
C. V3 ---+ X,B =} V3/B ---+ X, t
-AUX -AUX
D. V
2
---+ X =} V
2
---+ X, SEPREF O'E{ab, an, ..., ZU}
+0' +0'
A.3 Feature Cooccurrence Restrictions
I. +AC ---+ +MC
II.+MC ---+ +FIN
Rules of GFG 1137
A.4 Linear Precedence Rules
a. V < X
+MC
b. X < V
-MC
c. +TOP < X
d. x
2
< SEPREF
e. +NoM < +DAT
+NoM < +ACC
+DAT < +ACC
-FOCUS < +FoCUS
+PRONoUN < -PRONOUN
f. daB < V3
Bibliography
Ades, A.E. and M.J. Steedman (1982) On the Order of Words.
Linguistics and Philosophy 4, 517-558.
Akmajian, A., S. Steele, and T. Wasow (1979) The Category AUX in
Universal Grammar. Linguistic Inquiry 10, 1-64.
Anderson, S.R. and S. Chung (1977) On Grammatical Relations and
Clause Structure in Verb-Initial Languages. In P. Cole and J.M.
Sadock, eds., Grammatical Relations, Volume 8, Syntax and
Semantics, 1-25. Academic Press, New York.
Bach, E. (1962) The Order of Elements in a Transformational Grammar
of German. Language 38, 263-69.
Baker, C.L. (1978) Introduction to Generative-Transformational
Syntax. Prentice Hall, Englewood Cliffs, New Jersey.
Barlow, M., D. Flickinger, and I.A. Sag, eds. (1982) Developments in
Generalized Phrase Structure Grammar. Stanford Working
Papers in Grammatical Theory, Vol 2. Indiana University Linguis-
tics Club, Bloomington, Indiana.
Bear, John (1981) Gaps as syntactic features. Master's thesis, Univer-
sity of Texas, Austin, Texas. [also available from Indiana Univer-
sity Linguistics Club.]
Behaghel, Otto (1932) Die deutsche Syntax, Volume 4. Germanische
Bibliothek, Heidelberg.
Berwick, R.C. and A.S. Weinberg (1982) Parsing Efficiency, Computa.-
tional Complexity, and the Evaluation of Grammatical Theories.
Linguistic Inquiry 13, 165-191.
Bierwisch, Manfred (1971) Grammatik des deutschen Verbs, Studia
Grammatica 2. Akademie Verlag, Berlin.
Bresnan, J., ed. (1982) The Mental Representation of Grammatical
Relations. MIT Press, Cambridge, Massachusetts.
Bresnan, J., R.M. Kaplan, S. Peters, and A. Zaenen (1982) Cross-serial
Dependencies in Dutch. Linguistic Inquiry 13, 613-635.
169
170
Bibliography
Chomsky, N. (1957) Syntactic Structures. Mouton, The Hague.
Chomsky, N. (1965) Aspects of the Theory of Syntax. MIT Press,
Cambridge, Massachusetts.
Chomsky, N. (1970) Remarks on Nominalization. In R. Jacobs and
P. Rosenbaum, eds., Readings in English Transformational
Grammar. Ginn, Waltham.
Chomsky, N. (1977) On Wh-Movement. In P.W. Culicover, T. Wasow,
and A. Akmajian, eds., Formal Syntax. Academic Press, New
York.
Chomsky, N. (1981) Lectures on Government and Binding. Foris,
Dordrecht.
Culy , C. (1984) The Complexity of the Vocabulary of Bambara. Un-
published. [forthcoming in Linguistics and Philosophy.]
Diderichsen, P. (1962) Elementar Dansk Grammatik. Gyldendal, Copen-
hagen.
Drach, Erich (1963) Grundgedanken der deutschen
Darmstadt: Wissenschaftliche Buchgesellschaft.
Nachdr. der 3 Aufl. Frankfurt am Main 1940.]
Duden (1973) Grammatik der deutschen Gegenwartssprache. 3.
Bibliographisches Institut, Mannheim, Wien, Zurich. [Neu berb. u.
erw. Aufl. Bearb. von Paul Grebe unter Mitwirkung von Helmut
Gipper, Max Mangold, Wolfgang Mentrup und Christian Winkler.]
Earley, J. (1970) An Efficient Context-Free Parsing Algorithm.
Communications of the ACM 13, 94-102.
Etzensperger, J. (1979) Die Wortstellung der Deutschen Gegen-
wartssprache als Forschsmqeob ject, De Gruyter, Berlin, New York.
Flickinger, D. (1983) Lexical Heads and Phrasal Gaps. In Proceedings of
the Second West Coast Conference on Formal Linguistics. Stan-
ford Linguistics Association, Stanford, California.
Fourquet, J. (1959) Strukturelle Syntax und Inhaltbezogene Grammatik.
In Helmut Gipper, ed., Sprache, Schlussel zur Welt. Festschrift
fur Leo Weisgerber, 134-145. Schwann, Dusseldorf.
Fourquet, J. (1971) Satzgliedfolge und Satzverneinung als Probleme der
Struktur. In Fragen der strukturellen Syntax und der kontrastiven
Grammatik, 151-159. Schwarm, Dusseldorf.
Gawron, J.M. et al. (1982) The GPSG Linguistics System. In
Proceedings of the 20th Annual Meeting of the ACL in Toronto,
Ontario. ACL.
Gazdar, G. (1979) English as a Context-Free Language. Unpublished.
Gazdar, G. and G. Pullum (1982) Generalized Phrase Structure Gram-
mar: A Theoretical Synopsis. Indiana University Linguistics Club,
Bloomington, Indiana.
Satzlehre.
[Fotemech.
Bibliography
171
Gazdar, G. and G.K. Pullum (1982) Subcategorization, Constituent Or-
der and the Notion 'Head'. In M. Moortgat, H.v.d. Hulst and
T. Hoekstra, eds., The Scope of Lexical Rules, 107-123. Foris,
Dordrecht, Holland.
Gazdar, G. and G.K. Pullum (1982) Phrase Structure and Categorial
Syntax: A Select Bibliography of Recent Work. In M. Barlow, D.P.
Flickinger and LA. Sag, eds., Developments in Generalized Phrase
Structure Grammar, Volume 2, Stanford 'Working Papers in
Grammatical Theory. Indiana University Linguistics Club,
Gazdar, G. and G.K. Pullum (1982) Natural Languages and Context-
Free Languages. Lirujuietics and Philosophy 4,471-504.
Gazdar, G., and LA. Sag (1982) Passives and Reflexives in Phrase Struc-
ture Grammar. In J. Groenendijk, T. Janssen, and M. Stokhof,
eds., Formal Methods in the Study of Language. Proceedings for
the Third Amsterdam Colloquium, 131-152. Amsterdam: Math-
ematical Centre Tracts.
Gazdar, G., E. Klein, G.K. Pullum and I.A. Sag (1982) Coordina.te
Structure and Unbounded Dependencies. In M. Barlow, D.P. Flick-
inger and LA. Sag, eds., Developments in Generalized Phrase
Structure Grammar, Volume 2, Stanford Working Papers in
Grammatical Theory. Indiana University Linguistics Club,
Gazdar, G., G. K. Pullum, and l. A. Sag (1982) Auxiliaries and Related
Phenomena in a Restrictive Theory of Grammar. Language 5,8,
591-638.
Gazdar, G. (1982) Phrase Structure Grammar. In G. Pullum and
P. Jacobson, eds., The Nature of Syntactic Representation. Reidel,
Dordrecht, Holland, Holland.
Gazdar, Gerald (1982) Unbounded Dependencies and Coordinate Struc-
ture. Linguistic Inquiry 12, 155-184.
Geach, P.T. (1972) A Program for Syntax. In D. Davidson and
G. Harman, eos., Semantics of Natural Language. Reydel,
Dordrecht.
Hale, K. (1981) On the Position of Walbit'i in a Typology of the Baee,
Indiana University Linguistics Club, Bloomington, Indiana.
Heidolph et al. (1981) Grurul ziuje einer deuteclien Grammatik.
Akademie Verlag, Berlin.
Jackendoff, R.S. (1972) Semantic Interpretation in Generative
Grammar. MIT Press, Cambridge, Massachusetts.
Jackendoff, R.S. (1977) XSynta:t,' A. St'udy of Phrase Structure. MIT
Press, Cambridge, Massachusetts.
Jacobs, J. (1982) Syntax unci Semantik del' Neqaiion im Deutschen.
Fink, Munich.
172 Bibliography
Jacobson, P. and G. Pullum, eds. (1982) The Nature of Syntactic
Representation. Dordrecht: Reidel.
Jecklin, A. (1973) Untersuchungen zu den Satzbauplaenen gesprochener
Sprache. Francke, Bern.
Johnson, M. (1984) A GPSG Account of VP Fronting in German. Un-
published. [University of California at San Diego.]
Joshi, A.K. (1983) How Much Context-Sensitivity is Required to Provide
Reasonable Structural Descriptions: Tree Adjoining Grammars. In
D. Dowty, L. Karttunen, and A. Zwicky, eds., Natural Language
Processing: Psycholinguistic, Computational, and Theoretical
Perspectives. Cambridge University Press, Cambridge, England.
Joshi, A.K. and L.S. Levy (1980) Phrase Structure Trees Bear More Fruit
than You Would Have Throught. Unpublished. [Revised and ex-
panded version of a paper presented at the 18th Annual Meeting of
the ACL, University of Pennsylvania, Philadelphia, June 1980.]
Karttunen, L. and M. Kay (1983) Parsing in a Free Word Order Lan-
guage. In D. Dowty, L. Karttunen, and A. Zwicky, eds., Natural
Language Processing: Psycholinguistic, Computational, and
Theoretical Perspectives. Cambridge University Press, Cambridge,
England.
Karttunen, L. and S. Peters (1979) Conventional Implicature. In C.K.
Oh and D. Dinneen, eds., Presupposition, Volume 11, Syntax and
Semantics, 1-56. Academic Press, New York.
Kay, Martin (1979) Functional grammar. In C. Chiarello et al., ed.,
Proceedings of the Fifth Annual meeting of the Berkeley Linguis-
tics Society, 141-158. Berkeley Linguistics Society, Berkeley.
Keenan, E.L. and B. Comrie (1977) Noun Phrase Accessibility and
Universal Grammar. Linguistic Inquiry 8, 63-99.
Kilbury, J. (1984) A Modification of the Earley-Shieber Algorithm for
Direct Parsing of ID/LP Grammars. Unpublished. [Technische
Universitat Berlin.]
Kiparsky, P. (1966) Ueber den deutschen Akzent, Studia Grammatica 7.
Akademie Verlag, Berlin.
Klein, Ewan, and Ivan A. Sag (1982) Semantic Type and Control. In
M. Barlow, D. Flickinger and LA. Sag, eds., Developments in
Generalized Phrase Structure Grammar, Volume 2, Stanford
Working Papers in Grammatical Theory, 1-25. Indiana University
Linguistics Club, Bloomington, Indiana. [Also to appear in revised
form in Linguistics and Philosophy.]
Lakoff, G. (1965) On the Nature of Syntactic Irregularity. Doctoral
dissertation, Harvard University, Cambridge, Massachusetts.
Langendoen, D.T. and P. Postal. (1982) The Vastness of Natural Lan-
guages. Paper presented at the 1982 Annual meeting of the LSA
held at San Diego, California.
Bibliography
173
Lenerz, Jiirgen (1977) Zur Abfolge nominaler Satzglieder im Deutschen,
Volume 5, Studien zur deutschen Grammatik. Nan, Tiibingen.
Maling, J. and A. Zaenen (1980) A PSR Treatment of Unbounded
Dependencies in Scandinavian La.nguages. In P. Jacobson and
G. Pullum, eds., The Nature of Syntactic Representation. Reydel,
Dordrecht.
McCloskey, J. (1979) Trans formationsl Syntax and Model Theoretic
Semantics. Reidel, Dordrecht.
Montague, R. (1974) English as a Formal Language. In R.H. Thomason,
ed., Formal Philosophy, 188-221. Yale University Press, New
Haven and London.
Montague, R. (1974) The Propel' Treatment of Quantification in Or-
dinary English. In R.H. Thomason, ed., Formal Philosophy,
247-270. Yale University Press, New Haven and London.
Nerbonne, J. (1984) German Temporal Semantics: Three-Dimensional
Tense Logic and a GPSG Fraqment, Doctoral dissertation, Ohio
State University, Columbus, Ohio.
Nerbonne, J. (1982) German Impersonal Passivs: A Non-Structure-
Preserving Lexical Rule. In D. Flickinger, M. Macken, and
N. Wiegand, eds., Proceedings of the First West Coast Conference
on Formal Linguistics. Stanford University, Stanford, California.
Nerbonne, J. (1982) Phantoms and German Fronting: Poltergeist Con-
stituents. Paper presented at the Annual meeting of the LSA held
at San Diego, California, December 1982.
Perlmutter, D.M. (1983) On the Unnaccusative Hypothesis. In D.M.
Perlmutter, ed., Studies in Relat-ional Grammar, Volume 1. The
University of Chicago Press, Chicago, Illinois.
Pesetsky, D. (1979) Russian Alo'rphology and Lexical Theory. Doctoral
dissertation, MIT, Cambridge, Massachusetts.
Peters, P.S. and R.W. Ritchie (1973) On the Generative Power of Trans-
formational Grammars. Information Sciences 6, 49-83.
Pollard, C. and I. Sag (1983) Reflexives and Reciprocals in English: An
Alternative to the Binding Theory. In M. Barlow, D. Flickinger,
and M.T. Wescoat, eds., Proceedings of the Second West Coast
Conference on Formal Linguistics. Stanford University, Stanford,
California.
Postal, P.M. (1971) Oroee-Ooer-Phenomena, Academic Press, New York.
Piitz, H. (1975) tiber die Ssmta del' Pronominal jorni "es " im modernen
Deutsch. Studien zur deutschen Grammatik 3, Tiibingen.
Pullum, G.K. (1983) Context-Freeness and the Computer Processing of
Human Languages. In Proceedings of the 21st Annual Meeting of
the ACL in Cambritlqe, Maseacliueette. ACL.
Pullum, G.K. and D. Wilson (1977) Autonomous Syntax and the
Analysis of Auxiliaries. Language 53, 1-37.
174
Bibliography
Pullum, G.K. (1982) Free Word Order and Phrase Structure Rules. In
J. Pustejovsky and P. Sells, eds., Proceedings of the Twelfth An-
nual Meeting of the North Eastern Linguistic Society. North
Eastern Linguistic Society, Amherst Massachusetts.
Ross, J.R. (1967) Constraints on Variables in Syntax. Doctoral disser-
tation, MIT, Cambridge, Massachusetts.
Russell, G. (1983) Compound Verbs and Costituent Order in German.
Unpublished. [University of Sussex, England.]
Sag, L, and E. Klein (1982) The Syntax and Semantics of English Exple-
tive Pronoun Constructions. In M. Barlow, D. Flickinger and LA.
Sag, eds., Developments in Generalized Phrase Structure
Grammar, Volume 2, Stanford Working Papers in Grammatical
Theory, 92-136. Indiana University Linguistics Club, Bloomington,
Indiana.
Scaglione, A. (1981) The Theory of German Word Order from the
Renaissance to the Present. University of Minnesota Press, Min-
neapolis.
Selkirk, E. (1982) The Syntax of Words. MIT Press, Cambridge, Massa-
chusetts.
Sgall, P., E. Hajicova, and E. Benesova (1973) Topic, Focus and Genera-
tive Semantics. Taunus, Kronberg.
Shieber, S.M. (1984) Evidence Against the Context-Freeness of Natural
Language. Unpublished. [forthcoming in Linguistics and
Philosophy.]
Shieber, S.M., S.U. Stucky, H. Uszkoreit, J.J. Robinson (1983) Formal
Constraints on Metarules. In Proceedings of the 21st Annual Meet-
ing of the ACL in Cambridge, Massachusetts. ACL.
Shieber, S., H. Uszkoreit, F. Pereira, J. Robinson, M. Tyson (1983) The
Formalism and Implementation of PATR-II. In B.J.Grosz and M.E.
Stickel, ed., Research on Interactive Acquisition and Use of
Knowledge. SRI-International, Final Report 1894.
Shieber, S.M. (1984) Direct Parsing of ID/LP Grammars. Linguistics
and Philosophy 7, 135-154.
Simpson, J. (1983) Aspects of Warlpiri Morphology and Syntax. Doc-
toral dissertation, MIT, Cambridge, Massachusetts.
Simpson, J. (1982) Discontinuous Verbs and the Interaction of Morphol-
ogy and Syntax. In M. Barlow, D. Flickinger, and M.T. Wescoat,
eds., Proceedings of the Second West Coast Conference on Formal
Linguistics. Stanford Linguistics Association, Stanford, California.
Steedman, M. (1983) Dependency and Coordination the Grammar of
Dutch and English. Unpublished. [University of Ediburgh.]
Stucky, S. (1982) Linearization Rules and Typology. Proceedings of the
First West Coast Conference on Formal Linguistics 1, 60-70.
Bibliography 175
Stucky, Susan (1981) Word Order Variation in Makua: a Phraee Struc-
ture Grammar Analysis. Doctoral dissertation, University of Il-
linois, Urbana-Champaign.
Thiersch, C.L. (1978) Topics in German Syntax. Doctoral dissertation,
MIT, Cambridge, Massachusetts.
Thiersch, C. (1982) A Note on "Scrambling" and the Existence of VP.
Wiener Linguistische Gazette 27/28, 83-95.
Thompson, H. (1982) Handling Metarules in a Parser for GPSG. In
M. Barlow, D. P. Flickinger, and l.A. Sag, eds., Developments .in
Generalized Phrase Structure Granimar, Volume 2, Stanford
Working Papers in Grammatical Theory. Indiana University Lin-
guistics Club, Bloomington, Indiana.
Uszkoreit, H. (1979) A Metrical Approach to Lexical Stress in German.
Unpublished. [University of Texas at Austin.]
Uszkoreit, H. (1983) A Framework for Processing Partially Free Word
Order. In Proceedings of the 21st Annual Meetinq of the ACL in
Cmabridge, lvlassachusetts. ACL.
Uszkoreit, H. and S. Peters (1983) On Some Formal Properties of
Metarules. Unpublished. [to appear in Linguistics and
Philosophy.]
Uszkoreit, H. (1982) German Word Order in GPSG. In D. Flickinger,
M. Macken, and N. Wiegand, eds., Proceedinqs of the First West
Coast Conference on F01'1nal L-inguistics. Stanford University,
Stanford, Calif..
Uszkoreit, H. (1982) Topicalization in Standard German. Paper
presented at the Annual meeting of the LSA held at San Diego,
California. December 1982.
Von Stechow, A. (1978) Deutsche Wortstellung und
Montague-Grammatik. Sonderforschungsbereich 99 Linguistik,
Univ. Konstanz.
Welin, W.C. (1979) Studies in Compuiationul Ttxt Comprehension,
Volume Monograph 5 from the lnst. of Ling .. Univ. of Stockholm.
Wunderlich, D. (1984) Ueber die Arguruent.e des Verbs. Unpublished.
[Univ. Diisseldorf'.]
Wunderlich, D. (1983) On the Compositionality of German Prefix Verbs.
In R. Bauerle, C. Schwarze, and A. v. Stechow, eds., Meaninq, Use,
and Interpretation of Language, -152-165. de Gruyter, Berlin.
Wunderlich, D. (1983) Zur Syntax del' Praposit.ionalphrase im Deutschen.
Unpublished. [forthcoming in Zeitschrift fi.ir Sprachwissenschaft.]
Wurzel, W.U. (1970) Der Fremdwordakzent im Deutschen. Linguistics
65, 87-108.
Zaefferer, D. (1984) Fraqeauedruecke und Fra.qcn im Deuiechen, Fink,
Munich.
Author Index
Ades 150
Akmajian 71
Anderson 45
Bach 5
Baker 77
Barwise 121
Bear 75
Behaghel 22, 23
Benesova 5
Berwick 43, 131
Bierwisch 5, 108
Bresnan 43, 52, 130
Chomsky 32, 71, 72, 75, 107
Chung 45
Comrie 57
Culy 5
Diderichsen 75
Drach 24
Earley 131
Etzensperger 5
Flickinger 43, 77, 153
Gawron 131
Gazdar 5, 5, 31, 32, 33, 38, 41, 42, 45,
45, 52, 53, 55, 58, 71, 72, 75, 77,
82, 83, 129, 145, 158
Geach 150
Hajicova 5
Hale 3
Heidolph 5, 12, 55, 145, 159
Jackendoff 33, 107
Jacobs 5
Jecklin 50, 123
Johnson 150
Joshi 43
Kaplan 52
Karttunen 5, 121
Kay 5,52
Keenan 57
Kilbury 132
Kiparsky 91
Klein 31, 37, 38, 39, 53, 75
Lakoff 139
Lenerz 2, 5, 21, 54, 118, 121, 133, 137.
139, 141, 143, 145
Maling 44, 155
McCloskey 45
Montague 33, 38, 53
Nerbonne 55, 52, 148, 158, 150
Piitz 148
Perlmutter 139
Perry 121
Pesetsky 107
Peters 5, 35, 41, 42, 121, 130, 153
Pollard 77
Postal 139
Pullum 5, 5, 31, 32, 33, 38, 41, 43, 45
45, 52, 55, 58, 71, 77, 82, 129
177
178 Author Index
Ritchie 41
Robinson 42
Ross 21
Russell 52, 84, 85, 91, 97
Sag 31, 37, 38, 39, 45, 53, 55, 58, 71,
75, 77, 83, 158
Scaglione 5
Selkirk 107
Sgall 5
Shieber 5, 35, 42, 43, 45, 52, 131, 153
Simpson 5, 107, 108
Steedman 150
Steele 71
Stucky 5, 42, 129, 130, 145
Thiersch 5, 55
Thompson 31, 35, 42
Uszkoreit 5, 5, 35, 42, 91, 130, 145,
151, 152, 158, 153
Von Stechow 5, 24
Wasow 71
Weinberg 43, 131
Welin 75
Wilson 71
Wunderlich 5, 85, 91, 92, 157, 152
Wurzel 91
Zaefferer 5
Zaenen 44, 155
Subject Index
AC 60
ACC 117
AUX 65
BSE 68
DAT 117
FIN 60
FOCUS 115, 121, 130
INV 85
MC 60
NOM 117
PAS 68
PERF 66
PRN 119
PRONOUN 117
PSP 66
TOP 77
Accessibility hierarchy 57
Active 138
Adjacency 72
Adjuncts 84, 114, 121
Adverbial phrases 2,7,57, 145
Adverbs 25
Affix-hopping 71, 72
Agenthood 141
Agentive von-phrases 150
Agentive object 59
Agentive subject 150
Agentiveness 141
Agentless events 147
Agreement 52
Agreement features 52
Anaphora binding 5
Aspect 15
Aspectual auxiliary 17, 69
Assertion clauses 27, 60, 149
Assertion main clauses 75
Associativity 53
Atomic LP rules
137, 142
Auxiliary preposing 160
Auxiliary transformation 71
Auxiliary verbs 2, 6, 15, 65, 69, 75
117, 118, 119, 136,
Basic order 9
Basic rules 34
Benefactive 142
Binary feature system 48
Bounding 77
Bracket Erasure Convention 108
Branching clause structure 158
Branching structure 158
Breton 130
by-phrases 150
CAP 37
Case 138, 140, 142
Case features 2, 120
Case marking 120, 130, 133
Categorial information 161
Category-valued feature percolation 44
Category-valued features III
Category hood 138
Center of information 141
CF languages 32, 36
CF grammars 32
CF-PS grammars 32
CF-PS rules 32, 37, 129
CF-PS syntax 32
Chomsky hierarchy 5, 40, 41, 43
179
180 Subject Index
Clausal complement 136
Clausal constituent 24
Clausal objects 2, 7
Clausal subjects 2, 7, 135
Clause bounded ness 78, 153
Clause core 8,9, 15, 17,28,81, 134
Clause rules 143, 144
Cleft formation 24, 163
Complements 84, 99, 106, 114, 121,
129
Complex LP rule 117, 118
Compositionality 95
Compound formation 81, 105
Compound word stress 92
Compounds 97
Compounds of directional adverbs and
directional verbs 97
Conditionals 11
Configurational languages 3, 127
Configurational syntactic properties
Conjunctions 28
Constant growth 43
Constituent questions 27, 28, 75
Context-free 43
Context-free languages 5, 36
Context-free phrase structure grammars
32
Context-freeness of GPSG 42, 131
Context-freeness of natural language
41
Context-sensitive languages 36
Contraction 81, 87, 99
Contraction metarules 160
Contrastive focus 54
Contrastive stress 85
Control Agreement Principle 37, 52, 68
Conventional implicature 121
Crossed dependencies 151
Crossing dependencies 43
CS languages 36
Danish 76
do Gesei z der uiach.senden. Glieder 22
daB-clauses 133, 134, 141
Dative Shift 110
Declarative 39
Default Value Assignments 36
Definite NPs 21, 24, 120
Definiteness 4, 120, 122, 125, 162
Demonstrative proform 156
Denominal prefixes 103
Dependency Grammar 6
Derivational morphology 86
Detransitivization 110
Deverbal nouns 86
Direct object 18, 56, 113
Discontinous constituent 50
Discontinuous subject of psychological
verbs 56
Discontinuous verb phrases 49
Discourse information 121
Discourse role assignment 4, 125, 162
Discourse roles 50, 122, 150
Discourse situations 121
Discourse-determined ordering principles
2
DOBJ 18, 21, 24
Dominance 72
Double accusative verbs 120, 136
Double infinitives 160
Double modals 16, 65
Drach's Law 24, 75
Dutch 43
DVA 36
Earley algorithm 131
e8 147
Embedded phrase 26
Emphatic stress 85
English 10, 11, 15, 19, 23, 27, 28, 35,
37, 69, 70, 108, 122, 126, 127, 153,
157, 158, 161
English particle verbs 109
Equi 38,74
Essential variables 35, 42
Exception features 144
Experiencer 143
Expletive es 147, 148, 149, 150, 151
Extraction 57, 59, 78, 102, 104, 149,
151, 162
Extraposition 8, 28, 44, 126, 134, 147,
163
FCR 36,37
Feature bundle III
Feature cooccurrence restrictions 36,
60,73,77
Feature Default Instantiation Principle
83
Feature instantiation 77, 79, 131
Feature instantiation principles 36, 116
Feature system 111
Features 33,36,37,48, 60, 71, 73, 110,
117, 129, 132
Subject Index
FFC 36
Field of Verbal Arguments and Free Ad-
juncts 18
Finite Closure 36, 42, 43, 131, 146
Finite Closure Constraint 163
Finnish 6
Fixed order languages 138
Fixed word order 3
Flat aux node 71
Flat structure 411, 64, 66, 72, 98, 127,
145, 146, 158, 160
Flattening 44
Focus 4, 23, 24, 27, 54, 114, 119, 120,
121, 147
Focus accent 24
Focus particles 28
Focus raising 2, 7, 151, 152, 155, 162
Foot feature 36,71
Foot Feature Convention 36
Formal competence model 123
Free adj uncts 44
Free adverbial phrases 25
Free dative 142, 143
Free word order 6
Free word-order languages 38
Fronting 26, 27, 55, 75, 76, 77, 104,
107, 124, 148, 159
FUG 3,4,6
Functional application 38, 63, 74
Functional Sentence Perspective 4, 6
Functional Unification Grarr.mar 3
Future tense auxiliary 17, 69, 70
Gap feature 36
Gapping 76, 160, 162
GB 3,6
Generalized Phrase Structure Grammar
31
Generative Power of GPSG 39
Generative Transformational Grammar
32
Genitive NPs 120
Genitive objects 132, 133
German 42
GFG 48
Government and Binding 3
GPSG 3, 31, 32, 109, 120
Grammatical functions 2, 59, 64, 1:20,
125, 130, 133, 136, 13", 138, 140,
142, 161
haben 16,66
181
Head 129, 130
Head feature 52, 71, 73
Head Feature Convention 36, 52, 61,
65,67,68,83
Head feature percolation 36
Head-complement compounds 113
Heaviness principle 124
Heavy constituents 24, 211
Heavy-NP-Shift 21, 22, 122
Hewlett-Packard 131
HFC 36
Hungarian 19
ID rules 35, 51
(ID/LP) 1
ID/LP 1,2,3,6,44, 129, 163
ID/LP format 38
Idiomatic complement 107
Idioms 86, 106
IDR doubles 34, 35, 36, 37
IDR triples 36, 37
Illocutions 4
Immediate dominance 34,38
Immediate dominance rules 34
Immediate dominance/linear precedence
1
Imperatives 11
Impersonal passives 148
Incorporation 81, 87, 93, 99, 102, WI),
111
Indefinite noun phrases 21
Indefiniteness 122
Indexed languages 43
Indirect object 18, 56, 113
Indirect speech 11
Infinitival complements 28, 132
Infixation 91, 92
Inflectional morphology 125
Inkorporierung 87, 93
Inseparable prefix verbs 90, 91
Inseparable prefixes 89
Interjections 12
Intonation 85, 162
IOBJ 18, 21, 24
Island constraints 77
Japanese 19, 42, 130
Kleene star 14, 18, 72, 146
la.~8en-constructions 160
Left Dislocation 24, :28, 122, 163
182 Subject Index
Left- branching 69
Lexical component 163
Lexical Functional Grammar 3, 97, 120
Lexical Head Constraint 43, 163
Lexical Integrity Hypothesis 6, 108
Lexical rules 43, 99, 163
Lexicon 163
LFG 3, 6, 43, 52, 97, 109, 130
Liberation rules 43, 45, 46
Linear adjacency 87
Linear order 6, 119, 122
Linear precedence 34, 37, 38, 87
Long distance dependencies 44, 56, 76,
151, 154, 155
LP component 39, 116, 117, 127
LP conditions 116, 117
LP rules 3, 4, 37, 39, 45, 115, 143
Main clauses 8, 9, 60
Main verb 2, 17,69
Main verb analysis 71
Main-clause constituent questions 10
Major category 26
Makua 130, 145
Markedness 123
Metagrammar 31, 32, 34, 42
Metarule component 42
Metarules 35, 42, 43, 44, 144, 163
Metric structure 23
Metrical properties 122
Mildly context-sensitive 44
Mitteilungszentrum 141
Modal auxiliary 17,69
Modal verbs 2, 6, 15, 16, 65, 69, 70, 75
Model-theoretic semantics 32, 33
Montague grammar 6
Morphological features 135
Morphological rule 95, 99
Morphology 108
Multi pie extractions 155
Multiple-constituent fronting 156
Nested dependencies 151
Nesting 155, 160
Nesting constraints 155
New information 19, 20
Nominalizations 86, 93, 94
Non-CF languages 146
Non-structure-building rules
110, 162
Nonagentive subject 55, 59, 139, 141
Noncategory fronting 162
6, 109,
Nonconfigurational languages 3
Nonconfigurational syntactic properties
1
Nondefinite NPs 24
Nonfinite verb group 17
Nonfocus 24, 114
Nonseparated prefixes 100
Nonterminal symbols 33
Object 25, 54, 59
Old information 19, 20
Order of main and auxiliary verbs 48
Order of verbal complements and ad-
juncts 113
Ordering 4
Ordering freedom 161
Parentheticals 30
Parse time 131
Parsing algorithms 131
Partially free constituent order 154
Partially free order 2, 121
Partially free word order 1, 24, 125,
132
Particle Movement 109
Particle verbs 3
Passive 35, 90, 137, 138, 141, 150, 158
Passive auxiliary 17,69
Past-participle 90
PATR 52
Percolation 52, 68
Perfective aspect 16, 73, 90
Perfective auxiliary 73
Personal pronouns 24, 114, 119, 122,
150
Phantom category 158
Phonological heaviness 122
Phrasal heads 130
Position of the finite verb 9
Possessive 93
Post positions 130
Pragmatic constraints 26
Pragmatic principle 24
Pragmatics 163
Prague School 6, 141
Predicative phrases 132
Predicatives 25, 120
Prefix 82, 83, 110
Prefixation 81, 105
Preposition stranding 28, 78
Prepositional objects 132, 133
Prepositional phrase 29
Subject Index
Prepositions 130
Present-participle 15
Presuppositions 163
PRF 98
Primary stress 23, 89
Profligate languages 43
Pronominalization 150, 162
Pronouns 21, 123
Prosody 23
Protorules 74
Pseudo-cleft formation 163
Psychological verbs 55, 56, 57, 59, 140
Quantifier scope 4, 5
Raising 38,74,75
Recursive languages 36
Recursively enumerable languages 42
Recursively enumerable set 41
Redundancy rules 99
Regular expression 72
Relation-changing lexical rules 140
Relation-changing rules 43, UO, 152
Relation-changing syntactic rules 140
Relational Grammar 120, 138
Relative clause extraposition 28, 44,
127
Relevant information 20
Reported speech 11
Resumptive pronoun 147
Rhythm of the sentence 23
Right-branching 69
Romance languages 15, 72
Rule Extension Principle 61
Rule extension principles 35, 39
Satzglied 24
Scrambling 45, 62
Scrambling transformations 4
Second position of the verb 2, 15
Secondary topicalization 152
sein 16, 66
Selection features 72
Semantic component 74
Semantics 129, 163
Semantics of GPSG 38
Semantics of the auxiliary system 74
Semantics of VP and clause rules 53
Sensation verbs 143
Sentence ellipsis 160
Sentential complements 13:!
Sentential subjects 120
183
Separable prefix verbs 2, 163
Separable prefixes 3,6, 17,81
SEPREF 83, 98
Shared knowledge 23
Situations 121
Slash 77, 154
Slash-feature notation 76
Slash-introduction metarules 155
Slash-termination metarule 156
Slavic languages 4, 19
Specificity 4
Stack 155
Stack-val ued features 155
Stress 122
Strong generative capacity 39
Structure flattening 52, 87
Stylistic principles 123, 132, 162
Stylistic rules 1:!2
Stylistic transformations 4
Subcategorization
145
Subcategorization feature 83
Subcategorization frames 2, 132
Su bcategorization marker 33
Su bclauses 8
SUBJ 18, :!1, 24
Subject 18, 25, 54, 56, 59, 68, 113
33, 110, Ill, 140,
Subject cliticization 58
Subject-object asymmetries 53
Subjectless sentences 7
Subjectless verbs 2, 143, 144
Subjunctive 11, 12
Subordinate clauses 10, 13, 75, 149
Subordinate conjunction 13
Subordinate questions 10
Swedish 76
Swiss German 43
Syntactic category 2, 33, 71, 130, 133,
135, 140, 161
Sy ntactic ordering principles 2
Syntactic primitives 64, 120, 130, 137
Syntactic rules 1::!2
Sy ntactically discontinuous lexical units
108
TG 32,41
That-clause :!8
Thema.tic argument 130
Thematic roles 2, 59, 1l0, 111, 141,
142, 143, 161, 162
Theme 141
Tonal dynamics 23
184 Subject Index
Topic 4, 26, 54, 119
Topicalization 27, 28, 44, 47, 76, 100,
148, 149, 155, 156
Trace 79
Transformational Grammar 138, 144
Transitive verb phrases 158
Tree-adjoining languages 43
Turing machine power 5, 41, 42, 130
TVP 158
Two place relations 93
Typology 129
UG 52
Unification 36,37,53,68
Universal communicative strategy 127
Unmarked order 21, 24, 114, 120, 121,
124, 125, 132, 133, 141, 142
Unmarked Order Principle 2
Utterance modifiers 28
Valency 93,94,97
Verbpartikeln 81
Verbzusiitze 81
Verb adjuncts 81
Verb fronting 109
Verb particles 81
Verb-final 6,9, 13, 14, 15,75
Verb-initial 6,9, 11, 14,75
Verb-second 6, 9, 13, 14, 15, 18, 24, 54,
75
Verb-verb compounding 160
VSO language 158
W. languages 3
Warlpiri 3, 42, 108, 161
Warlpiri preverbs 108, 109
Weak generative capacity 39, 41, 161
Weakly context-free grammars 43
Weather verbs 147
werden 16
Wh-fronting 76, 155
Wh-Movement
153
Word boundaries 82
Word formation rules 99
Word order 2, 23, 162
Word Order Freedom 44, 127, 130
Word order variation 3
Word stress 92
28, 44, 76, 151, 152,
X languages 3
X -syntax 33, 48, 129
Order No. 12226-3

Word Order and Constituent Structure in German

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Word Order and Constituent Structure in German

Hochgeladen von

Copyright:

Verfügbare Formate

CENTER FOR THE STUDY

Das könnte Ihnen auch gefallen