Stylebook for Tübingen Treebank - Seminar für Sprachwissenschaft

Stylebook for the Tübingen Treebank
of Written German (TüBa-D/Z)
Heike Telljohann, Erhard W. Hinrichs, Sandra Kübler,
Heike Zinsmeister, Kathrin Beck
Universität Tübingen
Seminar für Sprachwissenschaft
Wilhelmstr. 19
D-72074 Tübingen
{telljohann,eh,kuebler,zinsmeis,kbeck}@sfs.uni-tuebingen.de
January 2012
Abstract
This stylebook is an updated version of Telljohann et al. (2009). It describes
the design principles and the annotation scheme for the German treebank
TüBa-D/Z developed by the Division of Computational Linguistics (Lehrstuhl
Prof. Hinrichs) at the Department of Linguistics (Seminar für Sprachwissenschaft – SfS) of the Eberhard Karls Universität Tübingen, Germany. The
guidelines focus on the syntactic annotation of written language data taken
from the German newspaper ’die tageszeitung’ (taz). The unannotated taz
newspaper material was taken from the Science CD (Wissenschafts-CD) of
’die tageszeitung’ (taz) that can be licensed from contrapress media GmbH
(http://shop.taz.de/index.php?cat=c18_taz-Archiv.html).
At present, the treebank comprises 65,524 sentences. The newspaper material
is taken from the taz editions from
1989 41 articles
1992 July 10, 11, 13, 14
1995 October 14, 16, 17
1997 237 articles
1999 April 30, May 3 – 7
The average sentence length is 17.8 words and the total number of tokens
currently amounts to 1,164,766. The TüBa-D/Z treebank is still under development. Thus, the number of annotated sentences will increase over time.
Periodic data updates and accompanying updates of this stylebook will be
made available at:
http://www.sfs.uni-tuebingen.de/en/tuebadz.shtml
Please consult this website in order to ensure that you are using the most
recent and most complete version of the treebank.
The annotation scheme for the TüBa-D/Z treebank is derived from the verbmobil treebank for spoken German, developed earlier (1997–2000) by the Division of Computational Linguistics of the SfS (Hinrichs et al. 2000). The
TüBa-D/Z annotation scheme has been extended along various dimensions
to accommodate the characteristics of written texts. In order to ensure the
reusability of the data, a surface-oriented annotation scheme has been adopted
that is inspired by the notion of topological fields and is enriched by a level of
predicate-argument structure. The linguistic inventory used in the treebank
annotation is based on a minimal set of assumptions that are uncontroversial
among major syntactic theories. In this sense it is an attempt at theoryneutrality.
i
Acknowledgements
Funding for the TüBa-D/Z has come from a variety of sources:
• the Competence Center for Text- and Information Technology (Kompetenzzentrum für Text- und Informationstechnologie – KIT)) grant by the
Ministry of Science, Research and the Arts Baden-Württemberg (funding
since 2000);
• the collaborative research center (Sonderforschungsbereich) grant SFB
441 – Linguistic Data Structures, project A1 – Representation and Automatic Acquisition of Linguistic Data funded by the German Research
Council (Deutsche Forschungsgemeinschaft – DFG);
• the collaborative research center (Sonderforschungsbereich) grant SFB
833 – The construction of meaning - the dynamics and adaptivity of linguistic structures, project A3 – Disambiguating Discourse Connectives
using Corpus-induced Semantic Relations funded by the German Research Council (Deutsche Forschungsgemeinschaft – DFG);
• the ESFRI research infrastructure project grants D-SPIN and CLARIND funded by the Federal Ministry of Education and Research (BMBF)
(funding since 2008).
A project of this scale would not be possible without the generous support
from many contributors:
Our special thanks go to ’die tageszeitung’ (taz) who kindly granted permission to process the newspaper data and to release the treebank.
We would like to acknowledge Rosmary Stegmann for her many contributions
to the treebank of spoken German in verbmobil. Her research laid the foundations for the annotation scheme of that treebank, which has been summarized in the ’Stylebook for the German Treebank in verbmobil’ (Stegmann
et al. 2000).
We would like to thank Manfred Sailer and Frank Richter for their helpful
comments and support in form of encouragement and critical discussions from
which we could strongly benefit for the challenging task of developing a dataoriented syntactic annotation scheme for spoken as well as for written German.
Furthermore, we are indebted to Tylman Ule for his assistance with part-ofspeech tagging of the data and with data conversion.
We would also like to acknowledge the support of Martina Liepert and Jorn
Veenstra, who initiated and developed the integration of named entities into
the annotation scheme.
Moreover, we would like to thank Julia Trushkina (Trushkina 2004) and Yannick Versley (Versley et al. 2010) who provided the tools for morphological
preprocessing.
ii
Furthermore, Yannick Versley (Versley et al. 2010) supported the project by
developing a tool for lemma disambiguation and for the automatic integration
of semantic classes of named entities.
The quality of the treebank has been considerably improved by feature oriented consistency checks developed by Ventsislav Zhechev. Further consistency tests were contributed by Tylman Ule and Frank H. Müller in the course
of their research work in the SFB 441. They deserve special mention for their
support.
We like to thank Vera Möller and Karin Naumann (2007) for annotating
anaphora and coreference relations and also for doing an excellent job in documenting the concepts.
Yannick Versley and Holger Wunsch supported the project in various aspects.
In the course of their Ph.D. projects in the SFB 441 they enhanced the conceptual aspects of the anaphora resolution as annotated in the treebank. They
also wrote mapping and conversion tools for integrating the anaphora annotion in the Export-XML format.
For their diligence and dedication to the arduous task of linguistic annotation and of post-editing we thank our research assistants Janne Berlacher,
Anne Brock, Armin Buch, Nadine Cetin, Marisa Delz, Silke Dutz, Katrin
Eichler, Emilia Ellsiepen, Steffen Froemel, Holger Gauza, Simone Hartung,
Daniel Hüttl, Heike Johannsen, Miriam Käshammer, Laura Kassner, Sarah
Klug, Janina Kopp, Anuschka Kranz, Christian Kreß, Rebecca Kreß, Michael
Kossack, Anne Lohse, Wolfgang Maier, Nicole Maruschka, Kai Metzger, Vera
Möller, Simone Müller, Maja Pietsch, Brigitta Rist, Andreas Rudin, Maria
Schmidt, Marie Schreier, Insa Starr, Melanie Störzer, Isabel Trott, and Dominikus Wetzel. They also improved the linguistic quality of the annotation
by dedicated discussions on problematic and interesting examples.
iii
The development of the TüBa-D/Z treebank was notably facilitated by a
number of former verbmobil partners whose contributions went well beyond
the call of duty. Hans Uszkoreit and his colleagues at the Saarland University
kindly provided us with the graphical annotation tool Annotate (Plaehn 1998)
which was developed as part of the research project (Teilprojekt C3; Principal investigators: Uszkoreit/Smolka) Nebenläufige grammatische Verarbeitung
(NEGRA) in the collaborative research center (Sonderforschungsbereich) 378.
The Annotate tool provides human annotators with a graphical, user-friendly
interface for annotating and editing trees and also offers database support for
maintaining large treebanks. We would like to express our special gratitude
to Thorsten Brants, who has kindly and generously provided us with software
support and user assistance for the Annotate tool from the very beginning of
the Tübingen treebank project.
iv
Contents
Abstract
i
Acknowledgements
iv
List of Tables
viii
1 Introduction
1
2 Major Challenges and Design Decisions
3
3 The Theoretical Basis of the Annotation Scheme
3.1 Topological Fields . . . . . . . . . . . . . . . . . .
3.1.1 The Concept of Topological Fields . . . . .
3.2 Constituent Analysis and Topological Fields . . . .
3.3 General Annotation Principles . . . . . . . . . . .
3.3.1 Flat Clustering Principle . . . . . . . . . .
3.3.2 Longest Match Principle . . . . . . . . . . .
3.3.3 High Attachment Principle . . . . . . . . .
3.4 The Structure of an Annotated Tree . . . . . . . .
3.4.1 The Levels of Annotation . . . . . . . . . .
3.4.2 The Inventory of Labels . . . . . . . . . . .
3.4.3 What Is a Syntactic Unit? . . . . . . . . .
3.4.4 Printing and Spelling Errors . . . . . . . . .
3.4.5 Isolated Phrases . . . . . . . . . . . . . . .
3.4.6 Long-Distance Dependencies . . . . . . . .
3.4.7 Empty Categories . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4 The Annotation of the Internal Structure of Phrases
4.1 Premodification and Postmodification in Phrases . . .
4.2 Noun Phrases . . . . . . . . . . . . . . . . . . . . . . .
4.2.1 Noun Phrases without Modifiers . . . . . . . .
4.2.2 Prenominal Modification . . . . . . . . . . . .
4.2.3 Postnominal Modification . . . . . . . . . . . .
4.2.4 Appositional Constructions . . . . . . . . . . .
4.2.5 Foreign Language Material . . . . . . . . . . .
4.2.6 Named Entity Annotation . . . . . . . . . . . .
4.2.7 Ordinal Numbers . . . . . . . . . . . . . . . . .
v
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
6
6
6
9
10
10
10
10
11
11
11
14
20
21
23
24
.
.
.
.
.
.
.
.
.
25
25
25
25
26
31
34
38
41
48
4.3
4.4
4.5
4.6
4.7
4.2.8 Cardinal Numbers . . . . . . . . . . . . . . . . . . . .
4.2.9 Letters and Non-Words . . . . . . . . . . . . . . . . .
4.2.10 Expletive and Other Uses of es . . . . . . . . . . . . .
Determiner Phrases . . . . . . . . . . . . . . . . . . . . . . .
Prepositional Phrases . . . . . . . . . . . . . . . . . . . . . .
4.4.1 Prepositions . . . . . . . . . . . . . . . . . . . . . . .
4.4.2 Circumpositions and Postpositions . . . . . . . . . . .
Adjectival Phrases . . . . . . . . . . . . . . . . . . . . . . . .
Adverbial Phrases . . . . . . . . . . . . . . . . . . . . . . . .
Verb Phrases . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.1 Head of a Sentence and Verb Complex . . . . . . . . .
4.7.2 Verb Complexes in Verb-second and Verb-final Clauses
4.7.3 Ersatzinfinitiv Constructions . . . . . . . . . . . . . .
4.7.4 Infinitives with zu . . . . . . . . . . . . . . . . . . . .
4.7.5 Coherency and Incoherency of Verbal Constructions .
4.7.6 AcI Constructions . . . . . . . . . . . . . . . . . . . .
4.7.7 Imperatives . . . . . . . . . . . . . . . . . . . . . . . .
4.7.8 Particle Verbs . . . . . . . . . . . . . . . . . . . . . .
4.7.9 Verbs with Predicate . . . . . . . . . . . . . . . . . . .
4.7.10 Modal Verbs . . . . . . . . . . . . . . . . . . . . . . .
5 Attachment Principles for Phrases
5.1 Attachment to Fields . . . . . . . . . . . . . .
5.2 Attachment of Ambiguous Complements . . . .
5.3 Modifier Attachment . . . . . . . . . . . . . . .
5.3.1 Modifier Attachment in the Initial Field
5.3.2 Attachment across Punctuation Marks .
5.3.3 Ambiguous Modifiers in Isolated Phrases
.
.
.
.
.
.
.
.
.
.
.
6 The Annotation of Sentences
6.1 Sentence Initial Fields . . . . . . . . . . . . . . . .
6.1.1 The C-Field in Verb-Final Clauses . . . . .
6.1.2 The C-Field in Verb-Second Clauses . . . .
6.1.3 The KOORD-Field in all Clause Types . .
6.1.4 The PARORD-Field in Verb-Second Clauses
6.1.5 Resumptive Constructions: The LV-Field .
6.2 Questions . . . . . . . . . . . . . . . . . . . . . . .
6.2.1 W-Questions . . . . . . . . . . . . . . . . .
6.2.2 Yes - No Questions . . . . . . . . . . . . . .
6.3 Relative Clauses . . . . . . . . . . . . . . . . . . .
6.3.1 Event-modifying Relative Clauses . . . . . .
6.3.2 Independent Relative Clauses . . . . . . . .
6.4 Coordination . . . . . . . . . . . . . . . . . . . . .
6.4.1 Coordination of Phrases . . . . . . . . . . .
6.4.2 Asymmetric Coordination . . . . . . . . . .
6.4.3 Coordinations with Complex Conjunctions
vi
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
48
50
51
54
55
55
58
58
63
65
65
65
67
69
71
72
73
74
75
78
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
80
80
80
81
83
83
84
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
86
86
86
88
88
89
90
91
91
91
92
94
94
95
96
97
98
.
.
.
.
.
.
.
.
.
.
.
6.5
6.4.4 Coordinations with Truncated Words . . . . . . . . .
6.4.5 Attachment Principles of Coordination within Phrases
6.4.6 Coordination of Topological Fields . . . . . . . . . . .
6.4.7 Attachment of Ambiguous Modifiers in Coordination .
6.4.8 Coordination of Sentences . . . . . . . . . . . . . . . .
6.4.9 Paratactic Constructions with denn and weil . . . . .
6.4.10 Conjunctions Occurring with Isolated Phrases . . . . .
6.4.11 Split Coordinations . . . . . . . . . . . . . . . . . . .
Elliptical Constructions . . . . . . . . . . . . . . . . . . . . .
7 The Annotation of Specific Syntactic Phenomena
7.1 Superlative and Comparative Forms . . . . . . . .
7.1.1 Superlative Forms . . . . . . . . . . . . . .
7.1.2 The Comparative Particles wie and als . . .
7.2 Verbal and Adjectival Use of Participles . . . . . .
7.3 Topicalization . . . . . . . . . . . . . . . . . . . .
7.4 Headlines . . . . . . . . . . . . . . . . . . . . . . .
7.5 Discourse Markers . . . . . . . . . . . . . . . . . .
7.6 Parentheses . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
8 Criteria for the Distinction of Grammatical Functions
8.1 Subcategorization of Verbs . . . . . . . . . . . . . . . .
8.2 Subcategorization of PREDs . . . . . . . . . . . . . . . .
8.3 Distinction of FOPP, OPP, and V-MOD . . . . . . . . .
8.4 Distinction of MOD, MOD-MOD, and V-MOD . . . . .
8.5 Distinction of ON, PRED, ON-MOD, and PRED-MOD .
9 The
9.1
9.2
9.3
9.4
9.5
TüBa-D/Z Data Formats
The NEGRA Export Format
The Penn Treebank Format
The Export-XML Format .
The TIGER-XML Format .
The CoNLL Format . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
99
101
102
103
105
107
107
108
109
.
.
.
.
.
.
.
.
112
112
112
112
115
116
117
119
121
.
.
.
.
.
123
123
123
124
125
125
.
.
.
.
.
128
128
132
135
137
139
References
140
Index
143
vii
List of Tables
3.1
3.2
3.3
3.4
3.5
3.6
3.7
3.8
3.9
Three clause types according to Höhle (1986) . . . . . .
Topological fields . . . . . . . . . . . . . . . . . . . . .
Levels of annotation . . . . . . . . . . . . . . . . . . .
The STTS tag set . . . . . . . . . . . . . . . . . . . . .
Morphological feature combinations for lexical elements
Values of morphological features . . . . . . . . . . . . .
Node labels . . . . . . . . . . . . . . . . . . . . . . . .
Edge labels . . . . . . . . . . . . . . . . . . . . . . . .
Syntactic-Semantic Node Labels for Named Entities . .
.
.
.
.
.
.
.
.
.
7
8
11
13
15
16
17
18
19
4.1
4.2
Semantic Classes and Subclasses for Named Entities . . . . . . . . . . . .
Types of es . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
42
53
viii
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Chapter 1
Introduction
The purpose of this report is to describe the design principles and annotation scheme for
the TüBa-D/Z treebank of German. It is intended as a guide for the treebank annotators
in Tübingen and for theoretical and computational linguists who want to use annotated
treebank data for their own research. In addition, we hope that this report may be
of some use for researchers who want to construct their own treebank for German or
for some other language. We would like to emphasize that the annotation scheme is
language-specific, and we advise against adopting this scheme without modification for
some other language. However, we do believe that the type of design decisions that are
reported here for German will arise for other languages as well. And it is in this sense
that the current report could provide an useful point of reference.
The TüBa-D/Z treebank was developed by the Division of Computational Linguistics
(Lehrstuhl Prof. Hinrichs) at the Department of Linguistics (Seminar für Sprachwissenschaft – SfS) of the Eberhard Karls Universität Tübingen, Germany. The guidelines focus on the syntactic annotation of written language data taken from the German newspaper ’die tageszeitung’ (taz). The unannotated taz newspaper material was taken from the
Science CD (Wissenschafts-CD) of ’die tageszeitung’ (taz) that can be licensed from contrapress media GmbH (http://shop.taz.de/index.php?cat=c18_taz-Archiv.html).
At present, the treebank comprises 65,524 sentences. The newspaper material is taken
from the taz editions from
1989 41 articles
1992 July 10, 11, 13, 14
1995 October 14, 16, 17
1997 237 articles
1999 April 30, May 3 – 7.
The average sentence length is 17.8 words and the total number of tokens currently
amounts to 1,164,766. The TüBa-D/Z treebank is still under development. Thus, the
number of annotated sentences will increase over time. Periodic data updates and accompanying updates of this stylebook will be made available at:
http://www.sfs.uni-tuebingen.de/en/tuebadz.shtml
Please consult this website in order to ensure that you are using the most recent and
most complete version of the treebank.
The annotation scheme for the TüBa-D/Z treebank is derived from the verbmobil
treebank for spoken German, developed earlier (1997–2000) by the Division of Compu1
tational Linguistics of the SfS (Hinrichs et al. 2000). The annotation scheme for the
verbmobil treebank has been summarized in the ’Stylebook for the German Treebank
in verbmobil’ (Stegmann et al. 2000). The TüBa-D/Z annotation scheme has been
extended along various dimensions to accommodate the characteristics of written texts.
In order to ensure the reusability of the data, the linguistic inventory used in the treebank annotation is based on a minimal set of assumptions that are uncontroversial among
major syntactic theories. In this sense it is an attempt at theory-neutrality.
The TüBa-D/Z treebank is released in five different data formats : the Negra Export
format, the Export-XML format, the TIGER-XML format, the Penn treebank format,
and the CoNLL format. More information about each data format is given in chapter 9.
To the best of our knowledge, the verbmobil treebank for spoken German is still
the only treebank based on non-genre-specific German speech data. It is released
as TüBa-D/S treebank (http://www.sfs.uni-tuebingen.de/en/tuebads.shtml). For
written texts, TüBa-D/Z is not the only treebank available for German. Two other
(semi-)manually annotated treebanks are currently available, each with their own annotation scheme: the Negra treebank (http://www.coli.uni-saarland.de/projects/
sfb378/negra-corpus/) and the TIGER treebank (http://www.ims.uni-stuttgart.
de/projekte/TIGER/).
The Tübingen Partially Parsed Corpus of Written German (TüPP-D/Z; http://www.
sfs.uni-tuebingen.de/en/tuepp.shtml) is a project closely related to the TüBa-D/Z
treebank. It consists of 200 million word tokens of the Science CD (Wissenschafts-CD)
of ’die tageszeitung’ (taz), including the sentences which are annotated in the TüBa-D/Z
treebank. The texts were automatically annotated with clause structure, topological
fields, and chunks, in addition to more low level annotation including parts of speech and
morphological ambiguity classes. The first release of TüBa-D/Z (12/2003) functioned as
training corpus.
2
Chapter 2
Major Challenges and Design
Decisions
Most syntactic theories consider individual sentences as the primary domain of linguistic
theorizing and of syntactic annotation. For written language, the segmentation into
sentences is largely unproblematic and coincides with the domain of syntactic analysis.
However, newspaper texts exhibit a number of phenomena that do not lend themselves
easily to a purely sentence-based annotation. These phenomena include: headlines, titles,
parentheses, discourse markers, and sentence conjunction by a colon. These cases are
described in more detail in sections 3.4.3 to 3.4.5 of this stylebook.
The second main question, which needed to be addressed at the outset of the project
was the inventory of syntactic categories and grammatical functions to be used for syntactic annotation and specification of predicate-argument structure. Here our choices were
guided by two main considerations:
1. Linguistic adequacy and theory-neutrality: For the purposes of reusability of
the treebank data, the annotation scheme should not reflect a commitment to a particular
syntactic theory. Rather, the inventory of categories should be a reflection of common
assumptions that syntacticians share across different frameworks concerning questions of
constituenthood, phrase attachment, and grammatical functions. On this note, the annotations should be theory-neutral and minimal. This desideratum is of utmost importance
so as to ensure the reusability of the annotated data.
At the same time, the annotation scheme should reflect as much as possible those
empirical generalizations that syntacticians, especially from a descriptive perspective,
have identified as characteristic of the language in question.
2. Balancing the needs of potential users: Since the construction of a treebank
is a labor-intensive and costly enterprise, ideally the TüBa-D/Z treebank should appeal
to as many potential users as possible. Moreover, the treebank should be of interest
to researchers of a wide range of different fields. Considering the renewed interest in
the use of corpora for both theoretical and computational linguistics, choicepoints in the
annotation scheme should be resolved in such a way that the needs of potential users are
balanced as much as possible.
3
To support the use of the TüBa-D/Z treebank in computational linguistics, the annotation scheme should be sensitive to processing considerations, as long as linguistic
adequacy of the choice of annotations is not compromised. Ceteris paribus, processing
considerations favor annotation schemes that pay close attention to properties of syntactic surface structure, particularly to word order regularities and distributional properties
of words and phrases. At the same time, the use of empty categories and data structures
with crossing dependencies among phrases are to be avoided if the annotations are to be
used for parsers that rely on the context-freeness of the underlying grammar.
In order to satisfy the above aims, the annotation scheme is surface-oriented and
context-free. The theoretical assumptions underlying the levels of annotation and the
choice of labels themselves are as much as possible based on a rich tradition of theoretical
and empirical research on German syntax.
For the treatment of word regularities of German, which is a language with relatively
free word order, an inventory of topological fields is incorporated into the annotation
scheme. Topological fields in the sense of Herling (1821), Erdmann (1886), Drach (1937),
and Höhle (1986) are widely used in descriptive studies of German syntax. Such fields
constitute an intermediate layer of analysis above the level of individual phrases and
below the clause level. The concept of topological fields favors tree-based annotations, i.e.
bracketings that do not rely on crossing or discontinuous dependencies. Instead, such nonlinear dependencies are to be expressed at the level of predicate-argument structure which
constitutes a second level of annotation with its own descriptive inventory of grammatical
functions.
The framework of topological fields is widely used in empirical and theoretical accounts
of German syntax. Thus, it is in the linguistics literature. This greatly facilitates thorough
training of human annotators, since they can rely on the pre-existing body of literature.
One purpose of this stylebook is to add to these reference materials.
Currently, a total of 25 syntactic node labels for the encoding of constituent structures
are being used. These include labels for topological fields as well as labels for phrases
and their constituent parts.
In order to capture grammatical functions of individual phrases and syntactic dependencies between phrases, constituent structure trees are enriched by a set of edge labels
between constituent structure nodes. The current inventory of edge labels comprises 42
distinct categories. In addition to these primary edge labels, four secondary edge labels are
used. These labels indicate phrase-internal government of elements in the verb complex,
express phrase-internal modification of noun phrases, resolve long-distance dependencies
among modifiers, or relate the phrasal complements of so-called third-construction control
verbs.
For certain computational applications, robust identification of named entities, e.g.
person names, names of companies and institutions, names of geographical locations, is
a major concern. Therefore, such named entities are identified by a special node label,
and their internal structure is sometimes identified by an additional secondary edge label
that is used exclusively for named entities.
At the word level, part-of-speech labels are assigned according to the Stuttgart-Tübingen tag set, which is widely accepted for part-of-speech tagging for German and which
provides an inventory of 54 distinct part-of-speech labels. In addition, information on
inflectional morphology is given.
4
Detailed information about the complete inventory of node labels, edge labels, partof-speech labels and inflectional feature clusters is given in section 3.4.2 of this stylebook.
The remainder of this stylebook is organized as follows: chapter 3 offers an overview
of the theoretical foundations of the annotation scheme, focusing on the concept of topological fields (3.1) and its relation to constituent structure (3.2), on general annotation
principles (3.3), as well as an overview of the annotation levels and of the inventory of the
annotation labels for each level (3.4). Chapter 4 concerns the annotation of the internal
structure of phrases, broken down into major word classes and their phrasal projections.
Chapter 5 addresses the principles for relating individual phrases to each other, particularly for modifier and complement attachment. Chapter 6 discusses the annotation
of entire sentences, focusing on the relationship between sentence types and topological
fields, coordination (including phrasal conjunction) and elliptical constructions. Chapter
7 is devoted to the annotation of miscellaneous syntactic constructions such as comparatives, verbal and adjectival participles, topicalization, newspaper headlines, discourse
markers, and parentheses, which each pose special challenges for the annotation tasks.
Chapter 8 describes the criteria used for distinguishing different grammatical functions.
Chapter 9 describes the five different data formats in which the TüBa-D/Z treebank is
distributed. The stylebook concludes with a bibliography and a subject index.
We do not consider the annotation level of anaphora and coreference relations in this
stylebook. Please consult (Naumann and Möller 2007) for a detailed description of these
phenomena.
5
Chapter 3
The Theoretical Basis of the
Annotation Scheme
3.1
Topological Fields
The annotation scheme for the TüBa-D/Z treebank has been developed with special
regard to the characteristics of the German language: the interaction of configurational
and non-configurational syntactic properties, which arise from the partially free word
order. On the one hand, there exist three different clause types with respect to the fixed
position of the finite verb (verb-second (V-2), verb-initial (V-1), and verb-final (V-end)).
On the other hand, there is a high degree of variability of complements and adjuncts. In
order to treat the relatively high degree of word order freedom in German, the treebank
adopts the notion of topological fields as the primary clustering principle of a sentence.
The basic characteristics of the model of topological sequences within a German sentence were originally formulated by Herling (1821) and Erdmann (1886). Herling (1821)
developed an adequate topological theory for complex sentences in which clauses form a
topological unit carrying a syntactic function and he mentioned the special position of the
finite verb in verb-second und verb-final clauses. Erdmann (1886) established the basics
of a theory of topological fields and pointed out that the first position in a clause is not
necessarily the subject position. The so called Herling/Erdmann scheme already covers
a set of word order regularities which apply for all three clause types of German. Later
Drach (1937) introduced the notion of field. Finally, Höhle (1986) developed topological
schemes for the three clause types.
3.1.1
The Concept of Topological Fields
In a German clause, the finite verb can appear in three different positions: verb-second,
verb-initial, and verb-final. Only in verb-final clauses the verb complex consisting of the
finite verb and non-finite verbal elements forms a unit. The discontinuous positioning
of the verbal elements in verb-first and verb-second clauses is the traditional reason for
structuring German clauses into fields. The positions of the verbal elements form the
Satzklammer (sentence bracket) which divides the sentence into a Vorfeld (initial field),
a Mittelfeld (middle field), and a Nachfeld (final field). The Vorfeld and the Mittelfeld
6
are divided by the linke Satzklammer (left sentence bracket), which is the finite verb,
the rechte Satzklammer (right sentence bracket) is the verb complex between the Mittelfeld and the Nachfeld. Thus, the theory of topological fields states the fundamental
regularities of German word order. It is an important basis for the topological analysis
of any German sentence, since subclauses and embedded clauses are treated within the
bounds of fields. Identical word order regularities within a specific field can be realized
in all three clause types. But the fields themselves differ in their possible elements and
grammatical rules. Therefore, the theory is a descriptive rather than explanatory theory
for a specific language.
Höhle (1986) denotes the three clause types as E-Sätze (verb-final clauses), F1-Sätze
(verb-initial clauses), and F2-Sätze (verb-second clauses). The topological schemes of
these types are listed in Table 3.1.
Table 3.1: Three clause types according to Höhle (1986)
E-Sätze (KOORD) - (C) - X - VK - Y
F1-Sätze (KOORD - (KL) - FINIT - X - VK - Y
F2-Sätze (KOORD or PARORD) - (KL) - K - FINIT - X - VK - Y
Abbreviations and explanations used in Table 3.1:
VK: verb complex
FINIT: element denoting categories of finiteness
KOORD: coordinating particles (e.g. und, oder)
PARORD: non-coordinating particles (e.g. denn, weil)
X, Y: sequence of any number of constituents
C: complementizer
K: one constituent
KL: nominativus pendens, resumptive construction (Linksversetzung)
These schemes topologically analyse not only atomic sentences but also complex sentence constructions which contain embedded clauses. Such embedded clauses can occur in
a Linksversetzung (resumptive construction), Vorfeld, Mittelfeld, or Nachfeld. Herling’s
theory of the coordination and embedding of sentences covers these phenomena in detail
(Herling 1821).
According to Höhle (1986), we assume the existence of the following topological fields
(cf. Table 3.2):
The following description of the topological fields does not claim completeness regarding all descriptive details but rather mentions their main characteristics.1
1
In the following, the abbreviations for the fields listed in Table 3.2 are used.
7
Table 3.2: Topological fields
Field
VF
LK
MF
VC
NF
LV
C
KOORD
PARORD
Description
Vorfeld (initial field)
Linke (Satz-)Klammer (left sentence bracket)
Mittelfeld (middle field)
Verbkomplex (verb complex)
Nachfeld (final field)
Linksversetzungsfeld (field for resumptive constructions)
C-Feld (field for complementizers, left from MF)
Koordinationsfeld (field for coordinating particles)
left-most element, optionally in all clause types, (e.g. und, oder)
Koordinationsfeld (field for non-coordinating particles)
left-most element, optionally only in verb-second (e.g. denn, weil)
VF: The Vorfeld consists of only one constituent. Usually it is the subject2 . But because
of the high degree of non-configurationality in German, the subject can also occur in the
Mittelfeld, thus allowing almost every other constituent to occupy the Vorfeld.
LK: The Linke Klammer is the position of the finite verb in verb-second and verb-first
clauses or a conjunction in verb-final clauses. It consists of exactly one element.
MF: Apart from those units which are optionally located in other fields, any nonverbal constituent may occur in the Mittelfeld. It consists of a sequence of any number
of constituents. The linear order of the constituents depends on the specific word order
principles for German and their interaction.
VC: The Verbkomplex is a sequence of verb forms. In verb-second and verb-first clauses
it consists of one or more non-finite elements or - depending on the verb - of a separable
prefix. In verb-final clauses it also contains the finite verb. The rule for the linear order
in general is: right determines left. If there is a finite verb in the verb complex, it is
usually the right-most element (exception: Ersatzinfinitiv constructions (daß er sich ein
neues Konzept wird überlegen müssen) (cf. 4.7.3).
NF: For some clause types (e.g. so daß-Sätze), the Nachfeld is the obligatory position.
Embedded complement clauses, relative clauses, and single constituents can optionally
occur in the Nachfeld. In contrast to the Vorfeld it may be occupied by any number of
constituents.
LV: The Linksversetzungsfeld is a field for the left-dislocated phrase of resumptive
constructions. A Linksversetzung is a pendent constituent. It can be regarded as a
2
In the fifth release, 52.5% of all Vorfeld fields host the subject.
8
syntactic anticipation of a part of a sentence (cf. 6.1.5). There are many restrictions
which apply for this position.
C: The C-Feld only occurs in verb-final clauses (exception: the conjunction als in subordinated sentences of comparison als wäre es nie geschehen.). It is obligatorily occupied
in finite verb-final clauses if there is no conjunction in the Linke Klammer. In non-finite
verb-final clauses the C-position may be empty. This field can be occupied by conjunctions of sentential objects (e.g. daß, ob) or sentence initial conjunctions like um, obwohl,
wenn and also by complex interrogative or relative phrases, e.g. ..., ’um wieviel Geld’ geht
es dabei? / ..., ’an der’ Max Daniel Professor für Klavier ist. (cf. 6.1.1).
KOORD: The KOORD-field is the field for coordinating particles. In contrast to the
PARORD-field, it can optionally occur as the left-most element of all clause types (cf.
6.1.3).
PARORD: The PARORD-field is the field for non-coordinating particles which optionally occur as the left-most element of a verb-second clause (cf. 6.1.4).
Concerning the distribution of constituents to topological fields see also the chapter
Deskriptive Generalisierungen in Grewendorf (1991).
The combination of these fields in order to constitute verb-first, verb-second, or verbfinal clauses is described in Höhle (1986).
The topological model, which is the basis of most traditional German grammars,
only provides descriptive parameters concerning the sentence structure without making
any statement about the regularities within the fields and the hierarchical constituent
structure of the sentence. For more complicated phenomena, it offers only a catalogue of
detailed descriptions.
3.2
Constituent Analysis and Topological Fields
The main weakness of the concept of topological fields is the above-mentioned fact that
the hierarchical constituent structure of a sentence cannot be described. The aim is to
find a form of representation which combines the topological model with a constituent
analysis in order to describe the hierarchy of the linguistic units within the fields. In our
annotation scheme, the integration of a constituent analysis was achieved by a second
level of annotation strictly within the bounds of topological fields: a predicate-argument
structure with its own descriptive inventory of syntactic categories and grammatical functions. The constituent structure is represented by phrase structure trees (phrase markers)
whose node and edge labels carry this information.
In order to analyse syntactic constructions, it is necessary to define the number and
types of constituents within the fields.
9
1. Number of constituents within the fields:
In general, C, LK, KOORD, PARORD, and VF contain only one constituent.
More than one constituent is allowed within MF and NF.
2. Types of constituents within the fields:
Phrasal constituents occur in VF, MF, NF and C (interrogative or relative phrases).
Embedded clauses either belong to NF, VF, LV, or in some cases to MF. Usually,
outside the spoken language context, verb-final clauses do not occur isolated. They
need to be attached if possible.
3.3
General Annotation Principles
Our annotation scheme tries to find a trade-off between pragmatic requirements on the
one hand and linguistic reality on the other hand. The following three common annotation
principles are adopted to group the constituents within a syntactic tree: the flat clustering
principle, the longest match principle, and the high attachment principle.
3.3.1
Flat Clustering Principle
The flat clustering principle keeps the number of hierarchy levels in a syntactic structure
as small as possible. As a consequence, any degree of branching is allowed. Constituents
which cannot be assigned a grammatical function within a syntactic construction are
structured as much as possible, but are not typically connected to surrounding constituents as a whole.
3.3.2
Longest Match Principle
The longest match principle demands that as many daughter nodes as possible are combined into a single mother node, provided that the resulting construction is syntactically
as well as semantically well-formed.
3.3.3
High Attachment Principle
The high attachment principle prescribes that syntactically and semantically ambiguous
modifiers are attached to the highest possible level in a tree structure. Premodifiers and
postmodifiers are treated in a different way. First, both kinds of modifiers are projected
to their phrase level. Since the modification scope of premodifiers is unambiguous, they
are directly attached to the head of the phrase which they are modifying. By contrast,
postmodifiers are always attached on a higher level to preserve ambiguity. This decision
was taken to avoid the problematic distinction whether a postmodifier is a free adjunct
or a complement of the modified phrase.
10
3.4
3.4.1
The Structure of an Annotated Tree
The Levels of Annotation
A syntactic tree consists of nodes and edges. Nodes represent constituents on different
levels of annotation. Edges always link daughter nodes to a mother node. The root node
of a tree is assumed as the sentence node of a construction. One level below the sentence
node, the nodes of the topological fields are located. This is the reason why topological
fields can be regarded as the top-level ordering principle for sentences in the treebank.
The sequence of the fields in the three clause types never violates the topological schemes
given by Höhle (1986). Within each sentence structure, in general at least two topological
fields are occupied (exception: infinitive constructions, (cf. 4.7.4). Others may be left
empty (elliptical constructions, cf. 6.5). Table 3.3 lists the four levels of annotation
which we distinguish within the structure of an annotated syntactic tree3 :
Table 3.3: Levels of annotation
Level
clause level
field level
phrase level
lexical level
Inventory
root node labels for different types of clauses
node labels for topological fields
(including labels for conjuncts of fields)
node labels for syntactic categories
(including syntactic-semantic node labels for named entities)
and edge labels for grammatical functions
lexical entries tagged with the part-of-speech (POS-)tags taken from
the STTS tag set (Schiller et al. 1995) and with morphological features
(Trushkina 2004, Versley et al. 2010) and lemmas (Versley et al. 2010)
Node labels denote the syntactic category of a phrase or sentence, a topological field,
or a grammatical property. Edge labels denote the grammatical function of lexical entries,
phrases, topological fields, and clauses.
3.4.2
The Inventory of Labels
The part-of-speech tags used for the annotation are taken from the Stuttgart-Tübingen
tag set (STTS) (Schiller et al. 1995).4 The STTS is a guideline for the annotation of
German text corpora on the lexical level. Every single part-of-speech of a text is assigned
one specific tag. The tag set consists of the tags listed in Table 3.4 (cf. (Schiller et al.
1995)). The tagging of the data was performed by the tnt tagger (Brants 1998) and
manually corrected with the Annotate tool (Plaehn 1998).
3
We do not consider the suprasentential annotation level of anaphora and coreference relations in this
stylebook. Please consult (Naumann and Möller 2007) for a detailed description of these phenomena.
4
PAV was changed into a new tag called PROP (pronominal form of a prepositional phrase) in order
to justify PX as the syntactic category of its mother.
11
The morphological tags give information about inflectional morphology and include
features such as case, number, person, etc. A specific combination of feature-value pairs is
defined for each relevant part-of-speech category, see Table 3.5 for the list of part-of-speech
categories that are annotated with morphological features and the corresponding feature
combinations. The values are represented in a cluster by single character abbreviations,
see Table 3.6 for the set of features and their values. Features can uniquely be identified
by their position in the cluster.
Node labels indicate the syntactic category of a phrase or sentence, but they are also
used to label topological fields and sequences of topological fields within coordinations or
to indicate specific grammatical properties of constituents. Table 3.7 lists all node labels
which are used in the treebank. (An additional node is introduced for named entities, see
Table 3.9)
Edge labels indicate the grammatical function of lexical entries, phrases, topological
fields, and clauses. Since case information is given and a distinction of different modifiers
is made by these labels, the syntactic tree structures also contain semantic roles. The
specific set of edge labels for the German treebank is listed in Table 3.8, including secondary edge labels. The latter ones are used to resolve ambiguities on a different level
of description.
Two specific edge labels denote whether a constituent has the function of a head
(HD), e.g. a phrase (NX, PX, ADJX, ADVX, VXFIN, VXINF), or a non-head (-), e.g. a
determiner or a modifier attached to a phrase. On any annotation level, there is at most
one head. Within phrases, these two labels indicate the internal dependency structure
of the phrase. The head of a sentence structure (e.g. SIMPX) is always the finite verb.
In coordinations, each conjunct depends on the head of the whole construction and is
denoted with a specific edge label (KONJ) in order to distinguish them from conjunctions
and modifying elements within a coordination (see 6.4.1 and 6.4.3). Edge labels below
all root node labels carry only non-head labels (cf. (Kübler and Telljohann 2002)).
In an enhanced version of the TüBa-D/Z treebank, each named entity is assigned one
of the following semantic classes: person (PER), organisation (ORG), location (LOC),
geopolitical entity (GPE), or other (OTH). The semantic class OTH comprises all remaining named entities not fitting into PER, ORG, LOC, or GPE (cf. 4.2.6).
In order to annotate these semantic classes, syntactic-semantic node labels of
the pattern syntactic category = semantic class are defined as the mother node of named
entities (see Table 3.9). These syntactic-semantic nodes indicate that the structure below
represents a (complex) named entity of a certain syntactic category belonging to one of
the five semantic classes (e.g. Ute Wedemeier (NX=PER), The Jim Wane Swingtett
(NX=ORG), Sögestraße (NX=LOC), Auf die stürmische Art (PX=OTH) (cf. 4.2.6).
The former node label ’EN-ADD’ and the secondary edge label ’EN’ are deleted.
The internal syntactic structure of named entities is governed by the general annotation rules. All parts below a syntactic-semantic node that do not belong to the named
entity itself are marked as ’-NE’, e.g. [[die (-NE)] AWO] (NX=ORG), [[Der (-NE)] zweite
Weltkrieg] (NX=OTH).
12
Table 3.4: The STTS tag set
POS =
ADJA
ADJD
ADV
APPR
APPRART
APPO
APZR
ART
CARD
FM
description
attributive adjective
adverbial or predicative adjective
adverb
preposition; left circumposition
preposition + article
postposition
right circumposition
definite or indefinite article
cardinal number
foreign language material
ITJ
KOUI
interjection
subordinating conjunction
with zu + infinitive
subordinating conjunction
with clause
coordinative conjunction
particle of comparison, no clause
noun
proper noun
substituting demonstrative
pronoun
attributive demonstrative
pronoun
substituting indefinite pronoun
attributive indefinite
pronoun without determiner
attributive indefinite
pronoun with determiner
irreflexive personal pronoun
substituting possessive pronoun
attributive possessive pronoun
substituting relative pronoun
attributive relative pronoun
reflexive personal pronoun
substituting interrogative pronoun
attributive interrogative pronoun
adverbial interrogative
or relative pronoun
pronominal adverb
KOUS
KON
KOKOM
NN
NE
PDS
PDAT
PIS
PIAT
PIDAT
PPER
PPOSS
PPOSAT
PRELS
PRELAT
PRF
PWS
PWAT
PWAV
PROP
13
examples
[das] große [Haus]
[er fährt] schnell, [er ist] schnell
schon, bald, doch
in [der Stadt], ohne [mich]
im [Haus], zur [Sache]
[ihm] zufolge, [der Sache] wegen
[von jetzt] an
der, die, das, ein, eine
zwei [Männer], [im Jahre] 1994
[Er hat das mit “]
A big fish [” übersetzt]
mhm, ach, tja
um [zu leben], anstatt [zu fragen]
weil, daß, damit, wenn, ob
und, oder, aber
als, wie
Tisch, Herr, [das] Reisen
Hans, Hamburg, HSV
dieser, jener
jener [Mensch]
keiner, viele, man, niemand
kein [Mensch], irgendein [Glas]
[ein] wenig [Wasser],
[die] beiden [Brüder]
ich, er, ihm, mich, dir
meins, deiner
mein [Buch], deine [Mutter]
[der Hund,] der
[der Mann ,] dessen [Hund]
sich, einander, dich, mir
wer, was
welche [Farbe], wessen [Hut]
warum, wo, wann, worüber, wobei
dafür, dabei, deswegen, trotzdem
POS =
description
examples
PTKZU
PTKNEG
PTKVZ
PTKANT
PTKA
TRUNC
VVFIN
VVIMP
VVINF
VVIZU
VVPP
VAFIN
VAIMP
VAINF
VAPP
VMFIN
VMINF
VMPP
XY
zu + infinitive
negation particle
separated verb particle
answer particle
particle with adjective or adverb
truncated word - first part
finite main verb
imperative, main verb
infinitive, main
infinitive + zu, main
past participle, main
finite verb, aux
imperative, aux
infinitive, aux
past participle, aux
finite verb, modal
infinitive, modal
past participle, modal
non-word containing
special characters
comma
sentence-final punctuation
other sentence internal punctuation
zu [gehen]
nicht
[er kommt] an, [er fährt] rad
ja, nein, danke, bitte
am [schönsten], zu [schnell]
An– [und Abreise]
[du] gehst, [wir] kommen [an]
komm [!]
gehen, ankommen
anzukommen, loszulassen
gegangen, angekommen
[du] bist, [wir] werden
sei [ruhig !]
werden, sein
gewesen
dürfen
wollen
[er hat] gekonnt
D2XW3, letters
$,
$.
$(
,
. ? ! ;:
-[]()
The following POS categories do not contain any morphological information and are
assigned the morphological label ”- -”: ADJD, ADV, APZR, CARD, FM, ITJ, KOUI,
KOUS, KON, KOKOM, PWAV, PROP, PTKZU, PTKNEG, PTKVZ, PTKANT, PTKA,
TRUNC, VVIZU, VVPP, VAPP, VMPP, XY, $, , $. , $( .
3.4.3
What Is a Syntactic Unit?
The newspaper articles of the taz have been defined as the primary segmentation domain
of the data. They are preprocessed into syntactic units delimited by punctuation marks
(. ? ! ; - ... /) for which specific rules demand or forbid segmentation. Each syntactic
unit is assigned a specific code which identifies its origin in the newspaper data, e.g.
T990507.123 (T (taz) 99 (year) 05 (month) 07 (day) 123 (article)).
A syntactic unit usually consists of one complete sentence structure with a root node
(SIMPX, R-SIMPX, P-SIMPX). But it may also consist of one or more sentences and/or
phrases, e.g. headlines, titles, sentences with parentheses, sentences with discourse markers, or sentence conjunction by a colon.
An annotated tree is a complete syntactically and semantically well-formed construction according to the longest match principle. The model of topological fields does not
prescribe that all fields have to be occupied. The fact that fields can be left empty, also
helps us to cope with elliptical constructions (cf. 6.5).
14
Table 3.5: Morphological feature combinations for lexical elements
POS
ADJA
feature combination
case number gender
APPR
case
APPRART
APPO
ART
NN
case number gender
case
case number gender
case number gender
NE
PDS
PDAT
PIS
case
case
case
case
PIAT
case number gender
PIDAT
case number gender
PPER
PWS
case number gender
person
case number gender
case number gender
case number gender
case number gender
case number gender
person
case number gender
PWAT
case number gender
PPOSS
PPOSAT
PRELS
PRELAT
PRF
number
number
number
number
gender
gender
gender
gender
comments
underspecified for gender if plural noun is underspecified, e.g. die/np* nordhessischen/np*
Grünen/np*
invariant local description e.g. Berliner/***
cardinal numbers as abbreviation: full morphology e.g. im 4./dsn Jahrhundert/dsn
without case if a preposition takes another
PP as complement, e.g. bis/ zu/d einer/dsf
Woche/dsf and in the construction was für
ein(er/e/...)
underspecified for gender, e.g. Abgeordnete (in
plural), Leute
underspecified: man/ns*
nichts/*** (cf. nix, sowas)
PIS or PIAT: allerhand/*** (cf. allerlei, allzuviel, dergleichen, derlei, etwas, genausoviel,
genug, genügend, keinerlei, mehr, reichlich,
soviel, viel, wenig, weniger, zuviel, zuwenig)
PIDAT or PIS: sowas/*** (cf. paar, bißchen)
plural is underspecified for gender, e.g.
lauter/***, see also ’PIS or PIAT’ below
solch/*** (cf. manch, welch, all), see also ’PIS
or PIDAT’ below
plural is underspecified for gender
sich: underspecified for gender
underspecified for gender: plural forms and
wer, wem, wen
wessen/***
15
POS
VAFIN
VAIMP
VMFIN
VVFIN
VVIMP
feature combination
person number mood
tense
number
person number mood
tense
person number mood
tense
number
comments
German has only second person imperative
forms
Table 3.6: Values of morphological features
Feature
case
gender
number
mood
person
tense
Values
n (nominative), g (genitive), d (dative), a (accusative), * (underspecified)
m (masculine), f (feminine), n (neuter), * (underspecified)
s (singular), p (plural), * (underspecified)
i (indicative), k (subjunctive; German ’Konjunktiv’)
1 (first), 2 (second), 3 (third), * (underspecified)
s (present), t (past)
Punctuation is not annotated, i.e., all punctuation marks are not attached to the tree
structure. Exceptions are punctuation marks which carry a semantic meaning within a
sentence, e.g. - (bis, und) in expressions like 15.30 - 17.30 Uhr. They are tagged according
to the part of speech that they represent in the text (cf. 4.4.1).
Constituents are not attached to a tree if they are not assigned a grammatical function within the specific syntactic construction. The following tree diagram shows two
annotated trees in one syntactic unit:5
5
These tree diagrams and all following tree diagrams in this report were generated with the aid of the
Negra Annotate tool.
16
Table 3.7: Node labels
Node Labels
ADJX
ADVX
DP
FX
NX
PX
VXFIN
VXINF
LV
C
FKOORD
KOORD
LK
MF
MFE
NF
PARORD
VC
VCE
VF
FKONJ
DM
P-SIMPX
R-SIMPX
SIMPX
Description
Phrase Node Labels
adjectival phrase
adverbial phrase
determiner phrase (e.g. gar keine)
foreign language phrase
noun phrase
prepositional phrase
finite verb phrase
non-finite verb phrase
Topological Field Node Labels
resumptive construction (Linksversetzung)
complementizer field (C-Feld)
coordination consisting of conjuncts of fields
field for coordinating particles
left sentence bracket (Linke (Satz-)Klammer)
middle field (Mittelfeld)
middle field between VCE and VC
final field (Nachfeld)
field for non-coordinating particles
verb complex (Verbkomplex)
verb complex with the split finite verb
of Ersatzinfinitiv constructions
initial field (Vorfeld)
conjunct consisting of more than one field
Root Node Labels
discourse marker
paratactic construction of simplex clauses
relative clause
simplex clause
17
Table 3.8: Edge labels
Edge Labels
Description
Edge Labels denoting Heads and Conjuncts
HD
head
non-head
KONJ
conjunct
Complement Edge Labels
ON
nominative object (i.e. subject; also clausal subjects)
OD
dative object
OA
accusative object
OG
genitive object
OS
sentential object
OPP
prepositional object
OADVP
adverbial object
OADJP
adjectival object
PRED
predicate
OV
verbal object
FOPP
facultative (i.e. optional) prepositional object,
passivized subject (von-phrase)
VPT
separable verb prefix
APP
apposition
Modifier Edge Labels
MOD
ambiguous modifier
ON-MOD, OA-MOD, OD-MOD, modifiers modifying complements or modifiers,
OG-MOD, OS-MOD, OPP-MOD, e.g. V-MOD = modifier of the verb
FOPP-MOD, PRED-MOD,
OADJP-MO, OADVP-MO,
V-MOD, MOD-MOD
Edge Labels in Split Coordinations
ONK, OAK, ODK, OGK,
second conjunct (K) in
OPPK, FOPPK,
PREDK, split coordinations
OSK, OADVPK, OA-MODK,
e.g. ONK = second conjunct
MODK, V-MODK
of a nominative object
Edge Label denoting Structural Expletive
ES
Vorfeld-es
Secondary Edge Labels
dependency relation between:
refvc
two verbal objects in VC
refmod
two ambiguous modifiers
refint
a phrase internal part and its modifier
refcontr
control verb and its complement
across clause boundaries
18
Table 3.9: Syntactic-Semantic Node Labels for Named Entities
Labels
Description
Syntactic-Semantic Node Labels
ADJX=ORG adjectival phrase, named entity of the semantic class “organisation”
ADJX=OTH adjectival phrase, named entity of the semantic class “other”
ADVX=ORG adverbial phrase, named entity of the semantic class “organisation”
ADVX=OTH adverbial phrase, named entity of the semantic class “other”
DM=OTH
discourse marker, named entity of the semantic class “other”
FX=LOC
foreign language phrase, named entity of the semantic class “location”
FX=ORG
foreign language phrase, named entity of the semantic class “organisation”
FX=OTH
foreign language phrase, named entity of the semantic class “other”
FX=PER
foreign language phrase, named entity of the semantic class “person”
NX=GPE
noun phrase, named entity of the semantic class “geopolitical entity”
NX=LOC
noun phrase, named entity of the semantic class “location”
NX=ORG
noun phrase, named entity of the semantic class “organisation”
NX=OTH
noun phrase, named entity of the semantic class “other”
NX=PER
noun phrase, named entity of the semantic class “person”
PX=GPE
prepositional phrase, named entity of the semantic class “geopolitical entity”
PX=LOC
prepositional phrase, named entity of the semantic class “location”
PX=ORG
prepositional phrase, named entity of the semantic class “organisation”
PX=OTH
prepositional phrase, named entity of the semantic class “other”
PX=PER
prepositional phrase, named entity of the semantic class “person”
SIMPX=ORG simplex clause, named entity of the semantic class “organisation”
SIMPX=OTH simplex clause, named entity of the semantic class “other”
VXINF=ORG non-finite verb phrase, named entity of the semantic class “organisation”
VXINF=OTH non-finite verb phrase, named entity of the semantic class “other”
Edge Label
-NE
non-head, the part below is not part of the named entity
19
SIMPX
511
−
−
−
−
VF
510
V−MOD
PX
506
−
LK
507
HD
HD
NX
500
−
An
MF
508
VXFIN
501
HD
HD
MOD
NX
502
HD
ADVX
503
HD
VXINF
504
HD
verwundet
NX
505
−
HD
der
Oder
er
dann
,
ein
Wadendurchschuß
APPR
ART
NE
VAFIN
PPER
ADV
VVPP
$,
ART
NN
$.
d
dsf
dsf
3sit
nsm3
−−
−−
−−
nsm
nsm
−−
0
1
2
wurde
VC
509
OV
ON
3
4
5
6
7
8
9
.
10
The leaves of the trees consist of pairs of non-terminal symbols and part-of-speech
tags. Non-terminal symbols are represented by spherical nodes, whereas edge labels are
depicted by rectangular nodes. The tree diagram consists of two trees, a SIMPX and
an isolated phrase. In accordance with the four annotation levels shown in Table 3.3,
the sentence is annotated top-down by the root node (SIMPX), the field nodes (VF, LK,
MF, and VC), the phrase nodes (PX, VXFIN, NX, ADVX, and VXINF), and finally
the tagged lexical entries. The edge labels between the field level and the phrase level
indicate that the syntactic structure contains one unambiguous modifier (V-MOD), a
subject (ON), one ambiguous modifier (MOD), a verbal object (OV), and the finite verb,
which itself is the head (HD) of the entire syntactic construction. The noun phrase (ein
Wadendurchschuß) is not attached to the sentence structure because otherwise the wellformedness of the construction would be violated. Thus, it has to be annotated as an
isolated phrase lacking a verbal constituent.
3.4.4
Printing and Spelling Errors
In contrast to spoken language data like in the Verbmobil (cf. (Stegmann et al. 2000))
which exhibit fragmentary utterances, false starts, repetitions, interruptions, and hesitation noises as its characteristic properties, data taken from newspaper corpora does not
include unintentionally formed syntactic constructions.
Deviations from syntactic wellformedness are either intended by the author or are
caused by printing errors. While incorrect writing of words is neglected in the syntactic
analysis (the respective lexical entry is marked with the correct writing of the word in a
comment line below), lexical elements which do not belong to the syntactic construction
(intentional or unintentional) are structured as much as possible, but are not attached to
the surrounding constituents:
20
−
SIMPX
511
−
−
−
MF
510
ON
VF
506
MOD
LK
507
HD
ADVX
500
HD
VXFIN
501
HD
Jetz
wollen
MOD
OA
NX
508
−
−
NX
502
HD
ADVX
503
HD
VC
509
OV
HD
ADJX
504
HD
VXINF
505
HD
Sie
wieder
ein
solches
ADV
VMFIN
PPER
ADV
ART
PIDAT
NN
VVINF
$.
−−
3pis
np*3
−−
asn
asn
asn
−−
−−
−
−
0
1
2
3
4
5
System
6
aufbauen
.
7
8
Jetzt
SIMPX
517
SIMPX
518
−
−
−
−
−
VF
515
V−MOD
NF
516
FOPP
PX
508
−
NX
500
HD
Am
LK
509
HD
VF
510
ON
LK
511
HD
MF
512
V−MOD
VXFIN
501
HD
NX
502
HD
VXFIN
503
HD
PX
504
HD
HD
0
Abend
1
erklärten
2
,
sie
3
4
seien
5
dabei
VC
513
6
PX
514
OV
HD
VXINF
505
HD
VXINF
506
HD
geschlagen
7
worden
−
HD
NX
507
−
8
−
von
9
10
HD
der
Polizei
11
APPRART
NN
VVFIN
$,
PPER
VAFIN
PROP
VVPP
VAPP
$(
APPR
ART
NN
dsm
dsm
3pit
−−
np*3
3pks
−−
−−
−−
−−
d
dsf
dsf
3.4.5
Isolated Phrases
There are textual fragments in newspaper data which cannot be analysed as a SIMPX
or as a constituent of a SIMPX because they are lacking a verbal constituent or they
are not assigned a specific grammatical function within a well-formed sentence. These
fragments are annotated as isolated phrases. The isolated elements are structured as
much as possible (mostly up to the level of phrasal categories), but they are not typically
connected to surrounding constituents as a whole, so that a conflict with the topological
field analysis is avoided. Their root node carries a phrasal category of their lexical head
(NX, PX, ADVX, etc.):
ADVX
503
−
−
PX
500
HD
Warum
ADVX
501
HD
0
auch
1
HD
ADVX
502
HD
nicht
2
?
3
PWAV
ADV
PTKNEG
$.
−−
−−
−−
−−
21
12
13
PX
503
−
HD
PX
502
−
HD
ADVX
500
HD
NX
501
HD
Hoffentlich
0
ohne
1
Nebenwirkungen
2
.
3
ADV
APPR
NN
$.
−−
a
apf
−−
In accordance with the longest match principle, as many parts of the fragment as
possible are projected to the phrase level and are included into a tree structure. It has
to be decided which part of the whole construction is the head and which parts depend
on this head.
Phrases within a syntactic unit are not attached on a higher level if they do not show
dependency relation. This is often the case with syntactic elements which are separated
by a colon or a dash (cf. 5.3.2):
SIMPX
NX
508
−
509
−
VF
−
HD
−
LK
505
NX
506
ON
HD
NX
VXFIN
500
507
−
VC
501
HD
ASB
504
HD
ein
1
:
2
HD
ADJX
503
VPT
lädt
0
NX
502
HD
−
HD
Tag
3
der
4
offenen
5
Tür
6
7
NN
VVFIN
PTKVZ
$.
NN
ART
ADJA
NN
nsm
3sis
−−
−−
nsm
gsf
gsf
gsf
NX
508
KONJ
NX
NX
500
−
501
−
Arlington
NX
HD
Road
0
502
,
3
R
NX
504
HD
1999
2
NX
503
HD
USA
1
NX
−
:
505
−
Mark
,
8
D
10
−
:
Jeff
11
507
−
−
Bridges
12
,
13
−
Tim
14
Robbins
4
5
6
NE
NE
NE
CARD
$,
NN
$.
NE
NE
$,
NN
$.
NE
NE
$,
NE
NE
nsn
nsf
npm
−−
−−
nsf
−−
nsm
nsm
−−
npm
−−
nsm
nsm
−−
nsm
nsm
22
9
NX
506
HD
Pellington
7
KONJ
NX
15
16
SIMPX
512
−
−
−
VF
510
V−MOD
MF
511
ON
ADVX
507
NX
500
HD
Berlin
NX
501
HD
HD
−
ADVX
502
HD
ADVX
503
HD
PX
509
−
VXFIN
504
HD
NX
505
HD
(
taz
)
−
So
NE
$(
$(
ADV
ADV
VAFIN
PIS
APPRART
NN
$.
nsn
−−
nsf
−−
−−
−−
−−
3sis
ns*
dsm
dsm
−−
3.4.6
2
3
4
5
6
man
NX
506
HD
$(
1
wird
HD
NE
0
also
PRED
LK
508
HD
7
zum
8
9
Problemfall
10
.
11
Long-Distance Dependencies
Our annotation scheme facilitates a surface-oriented representation of long-distance dependencies without crossing branches and traces. If a modifying constituent is not adjacent to the modified constituent, their dependency relation, which can even go beyond the
border of topological fields, is encoded by special naming conventions for edge labels. We
use edge labels such as OA-MOD (referring to OA) or PRED-MOD (referring to PRED)
etc. expressing the non-ambiguity of the modifier.
Beyond this, we make use of secondary edge labels for ambiguity resolution. These labels just serve as additional information to the grammatical functions encoded in the edge
labels. These secondary edge labels indicate underspecified long distance dependencies
in the following cases:
1. If the above mentioned edge labels need further disambiguation, e.g. if there are
two OAs or V-MODs below one SIMPX node (refmod).
2. If the dependency relation exists between two nodes of which at least one is phrase
internal and therefore carries only head or non-head information (refint).
3. If there is a dependency relation outside of SIMPX in control verb constructions
(refcontr).
SIMPX
512
−
−
−
VF
507
ON
LK
508
HD
V−MOD
MF
509
MOD
NX
500
HD
VXFIN
501
HD
ADVX
502
HD
ADJX
503
HD
Die
ADJX
refmod
504
HD
−
je
.
VVINF
KOKOM
ADV
$.
np*
3pis
−−
−−
−−
−−
−−
−−
−−
5
denn
HD
ADJD
4
schlummern
ADVX
506
ADJD
3
seliger
VXINF
505
HD
ADV
2
künftig
V−MOD
506
VAFIN
1
dort
−
NF
511
MOD−MOD
PDS
0
werden
−
VC
510
OV
23
6
7
8
−
SIMPX
515
−
−
−
MF
513
OA
NF
514
MOD
NX
511
SIMPX
512
HD
VF
506
ON
LK
507
HD
NX
500
HD
VXFIN
501
HD
Dieser
hat
−
−
PX
508
refint
−
HD
512
NX
502
HD
NX
503
−
NX
504
HD
VXINF
505
HD
die
Bereitschaft
,
Therapieangebote
VAFIN
NN
APPR
ART
NN
$,
NN
VVIZU
$.
nsm
3sis
apf
a
asf
asf
−−
apn
−−
−−
1
−
auf
VC
510
HD
PDS
0
Auswirkungen
HD
−
MF
509
OA
2
3
4
SIMPX
512
−
−
5
6
7
anzunehmen
8
.
9
−
NF
511
OS
SIMPX
510
−
VF
505
OA
LK
506
HD refcontr
NX
500
−
MF
507
ON
VXFIN
501
HD
HD
NX
503
−
−
das
zu
schicken
VVFIN
PIS
ART
NN
PTKZU
VVINF
$.
***
asn
3sks
ns*
dp*
dp*
−−
−−
−−
3.4.7
3
4
Angehörigen
HD
PDS
2
den
VXINF
504
HD
All
1
man
VC
509
HD
500
PIDAT
0
versuche
NX
502
HD
−
MF
508
OD
5
6
7
.
8
Empty Categories
In general, an empty category analysis, e.g. for phrases without heads, is being avoided
in the TüBa-D/Z treebank.
Empty Edge Labels
Specifiers, prepositions,6 complementizers, discourse markers, KOORD and PARORD
constituents, conjunctions, and unambiguous modifiers (that are attached to phrases immediately rather than to topological fields ) are not labelled with grammatical functions.
Furthermore, the edges below the SIMPX node are empty. They are not labelled in order
to speed up annotation where the information is unnecessary or self-evident.
Furthermore, empty edge labels are used in elliptical phrases, e.g. noun phrases only
consisting of an article and an attributive adjective (cf. 6.5).
6
In order to facilitate the identification of dependencies between verbs and their nominal complements
and adjuncts and in keeping with basic assumptions in Dependency Grammar, the annotated head of a
prepositional phrase is the NX (or complement) rather than the preposition itself. Therefore, prepositions
carry no edge label.
24
Chapter 4
The Annotation of the Internal
Structure of Phrases
4.1
Premodification and Postmodification in Phrases
The annotation of phrases is also carried out following the flat clustering principle in
order to keep the number of hierarchy levels in a syntactic structure as small as possible.
As will be shown in the following sections, phrases may include adjectival or nominal
premodifiers and/or postmodifiers of any syntactic category. Both kinds of modifiers are
in principle projected to their phrase levels. Since the modification scope of premodifiers
is unambiguous, they are directly attached to the head of the phrase which they modify.
By contrast, postmodifiers are always attached on a higher level to preserve ambiguity.
This decision, referred to in 3.3 as the high attachment principle, was made to avoid the
problematic distinction whether a postmodifier is a free adjunct or a complement of the
modified phrase. The attachment strategy for premodifiers and postmodifiers is applied
for all categories of phrases.
4.2
Noun Phrases
A simple noun phrase (NX) consists of a head noun (noun, proper noun, or a pronoun),
(optionally) a determiner and (optionally) an adjectival or a nominal premodifier of any
complexity preceding the head noun. A complex noun phrase is a simple noun phrase
with a postmodifier of any syntactic category and complexity.
4.2.1
Noun Phrases without Modifiers
Simple noun phrases without modifiers are single nouns, proper nouns, pronouns or proper
nouns consisting of more than one NE. All of them are directly projected to their phrase
level. While single nouns, proper nouns and pronouns carry the edge label HD, the NEtagged tokens of a complex proper noun are attached on the same level without head
information:
25
NX
500
HD
Spendengeld
0
NN
asf
NX
500
HD
Hamburg
0
NE
dsn
NX
500
−
−
Ute
Wedemeier
0
1
NE
NE
nsf
nsf
If proper nouns include other parts of speech than NEs, these parts are tagged according to their distribution. Therefore, proper nouns with a preposition include a prepositional phrase.
NX
502
−
−
PX
501
−
HD
NX
500
HD
Ole
0
von
1
Beust
NE
APPR
NE
nsm
d
dsm
4.2.2
2
Prenominal Modification
In a simple noun phrase, both the determiner and the head noun are directly attached
on the same level to NX so that the label of the head noun carries the edge label HD and
the edge label of the determiner is empty.
26
NX
500
−
HD
die
Auseinandersetzung
0
1
ART
NN
nsf
nsf
NX
500
−
HD
jede
Spur
0
1
PIDAT
NN
nsf
nsf
Since prenominal modifiers are directly attached to the head noun on the same level,
their edge labels are empty (whereas the edge labels of modifiers that are attached to
topological fields are non-empty (cf. 8.4)). Prenominal modifiers are either attributive
adjectives or preceding genitive phrases:
NX
501
−
−
HD
ADJX
500
HD
ein
externer
0
Wirtschaftsprüfer
1
2
ART
ADJA
NN
nsm
nsm
nsm
NX
501
−
−
HD
ADJX
500
HD
−
die
zu
verhandelnden
ART
PTKZU
ADJA
NN
npf
−−
npf
npf
0
1
2
Taten
3
27
NX
501
−
HD
NX
500
HD
Bremens
Gesundheitssenatorin
0
1
NE
NN
gsn
nsf
If there is a PIDAT preceding the article it is directly attached to the noun phrase.
NX
501
−
−
−
HD
ADJX
500
HD
all
die
historischen
PIDAT
ART
ADJA
NN
***
apm
apm
apm
0
1
2
Fehler
3
If a PIDAT is following the article in adjective position it is projected to its phrase
level (ADJX) with possible premodifiers and then directly attached like an attributive
adjective to the noun phrase.
NX
501
−
−
HD
ADJX
500
HD
Die
meisten
0
Benutzer
1
2
ART
PIDAT
NN
npm
npm
npm
28
NX
504
−
−
HD
ADJX
503
−
−
HD
PX
502
−
HD
NX
500
ADVX
501
HD
die
in
0
HD
Deutschland
1
ohnehin
2
wenigen
3
Gen−Food−Produzenten
4
5
ART
APPR
NE
ADV
PIDAT
NN
npm
d
dsn
−−
npm
npm
If there is more than one prenominal modifier, the one on the left hand side of the
noun is modifying the following noun, the one on the left hand side of the modifier is
modifying both, the modifier and the noun, and so on. All of these modifiers are attached
to the head noun on the same level which yields a rather flat noun phrase structure.
This strategy is justified by the fact that these modifiers have a scope of modification
beyond the adjectival phrase, e.g. as in coordinated noun phrases like insgesamt 12.000
Studienplätze und 15.000 Lehrstellen, the adverb insgesamt modifies 12.000 Studienplätze
as well as 15.000 Lehrstellen.
NX
502
−
−
ADJX
500
HD
ADJX
501
HD
HD
lieber
knieartiger
0
Leser
1
2
ADJA
ADJA
NN
nsm
nsm
nsm
In case of complex head nouns, e.g. complex (proper) nouns consisting of two nominal
parts or coordinated head nouns (cf. 6.4.5), first the complex noun respectively the coordination (cf. 6.4) is annotated with its own internal dependency structure. Afterwards,
the determiner and possible premodifying adjectival phrases are attached on a higher
level.
29
NX
502
−
HD
NX
501
−
HD
NX
500
−
Die
"
0
−
debis
1
Systemhaus
2
GmbH
3
"
4
5
ART
$(
NE
NE
NN
$(
nsf
−−
nsf
nsn
nsf
−−
NX
502
−
HD
NX
501
−
HD
NX
500
HD
der
Heinrich
0
Böll−Stiftung
1
2
ART
NE
NN
gsf
gsm
gsf
NX
504
−
HD
NX
503
HD
−
PX
502
−
HD
NX
500
HD
NX
501
HD
"
Solidarität
$(
NN
APPR
−−
nsf
d
0
1
mit
2
Miloevic
"
−
Parolen
NE
$(
$(
NN
dsm
−−
−−
dpf
3
4
5
Milosevic
30
6
NX
503
−
HD
NX
502
KONJ
ihren
−
KONJ
NX
500
NX
501
HD
HD
Sänger
0
und
1
Gründer
2
3
PPOSAT
NN
KON
NN
asm
asm
−−
asm
4.2.3
Postnominal Modification
Whereas prenominal modifiers are always directly attached to the head noun on the same
level, postnominal modifiers are attached to the head noun on a higher level. Postnominal
modifiers are also always first projected to the phrase level before they are attached to
the head noun on a higher level. Phrase internal postmodifiers can be of any phrasal
category. The following tree structures show a prepositional phrase (PX) and a genitive
phrase (NX) as postmodifiers. See section 6.3, page 92 for the analysis of relative clauses.
NX
503
HD
−
PX
502
−
HD
NX
500
NX
501
HD
HD
Glück
im
0
Netz
1
2
NN
APPRART
NN
nsn
dsn
dsn
NX
503
HD
−
NX
502
−
−
NX
500
−
ADJX
501
HD
Die
HD
HD
Mitteilung
0
des
1
Bremer
2
Senats
3
4
ART
NN
ART
ADJA
NN
asf
asf
gsm
***
gsm
In case a noun has more than one postmodifier, these modifiers usually show a hierarchical structure, for example, the first modifier modifies the head noun, the second
31
modifier modifies the complete preceding noun phrase structure, and so on.
NX
506
HD
−
NX
505
HD
−
NX
503
−
PX
504
−
HD
−
ADJX
500
HD
die
guten
0
Beziehungen
NX
501
NX
502
HD
HD
Bonns
1
HD
2
zu
3
Moskau
4
5
ART
ADJA
NN
NE
APPR
NE
apf
apf
apf
gsn
d
dsn
Attributes of degree and quantity nouns are also defined as postnominal modifiers:
NX
502
HD
−
NX
500
NX
501
−
HD
eine
HD
Kiste
0
Sprengstoff
1
2
ART
NN
NN
asf
asf
asm
Cardinal numbers either appear as quantity nouns or premodifying adjectival attributes, e.g. the cardinal number 1,000,000 can also be expressed by the quantity noun
eine Million. Therefore, we have to distinguish the following two ways of annotation:
SIMPX
510
−
−
VF
509
ON
NX
508
HD
−
PX
507
−
HD
NX
506
HD
−
NX
504
−
NX
500
−
Der
ADJX
501
HD
HD
0
Etat
LK
505
HD
HD
1
von
2
3,5
3
NX
502
HD
Millionen
4
Mark
5
VXFIN
503
HD
steht
6
.
7
ART
NN
APPR
CARD
NN
NN
VVFIN
$.
nsm
nsm
d
−−
dpf
dpf
3sis
−−
32
SIMPX
517
−
−
−
VF
516
OS
SIMPX
515
−
−
−
VF
513
ON
MF
514
MOD
NX
508
HD
−
NX
500
−
NX
501
HD
HD
NX
510
−
VXFIN
502
HD
NX
506
HD
NX
507
HD
Das
5
Mark
"
,
empört
NN
NN
VVFIN
ADV
CARD
NN
$(
$,
VVFIN
PPER
PRF
$.
−−
nsn
nsn
nsn
3sit
−−
−−
apf
−−
−−
3sis
nsf3
as*3
−−
3
zuletzt
VXFIN
505
HD
OA
ART
2
kostete
ADJX
504
HD
MF
512
ON
"
1
Weißbrot
ADVX
503
HD
LK
511
HD
HD
$(
0
Kilo
OA
LK
509
HD
4
5
6
7
8
9
10
sie
11
sich
12
.
13
For nominal postmodifiers apart from genitive phrases the same attachment rule is applied. This kind of postmodifiers which may also appear in brackets, e.g. Heinz Schleußer
(SPD), is semantically closely related to the preceding head noun phrase. die Arbeiterwohlfahrt Bremen, for instance, means die Arbeiterwohlfahrt which is located in Bremen,
but does not mean die Arbeiterwohlfahrt which is called Bremen. Hence, these postmodifiers have to be distinguished from appositions (cf. 4.2.4) and complex named entities
(cf. 4.2.6).
NX
502
HD
−
NX
NX
500
501
−
−
Heinz
HD
Schleußer
0
(
1
SPD
2
)
3
4
NE
NE
$(
NE
$(
nsm
nsm
−−
nsf
−−
NX
502
HD
−
NX
NX
500
−
501
HD
die
HD
Arbeiterwohlfahrt
0
Bremen
1
2
ART
NN
NE
nsf
nsf
nsn
33
NX
502
HD
−
NX
500
HD
NX
501
HD
Zentralkrankenhaus
0
Ost
NN
NN
dsn
dsn
1
NX
502
HD
−
NX
500
HD
Kapitel
NX
501
HD
0
VII
1
NN
CARD
dsn
−−
NX
502
HD
−
NX
500
−
des
NX
501
HD
HD
0
ICE
1
884
2
ART
NN
CARD
gsm
gsm
−−
4.2.4
Appositional Constructions
An apposition is a specific kind of attribute to a noun, which normally agrees in case
with this noun and does not change its overall meaning. There is no consensus among
grammarians of what is exactly meant by the notion apposition (cf. (Eisenberg 1999
2001)). Eisenberg (1999 2001), for instance, claims that, e.g. Ute Wedemeier die Landesvorsitzende and die Landesvorsitzende Ute Wedemeier are both appositions but it is
not clear which part is the apposition and which part is the head noun. The Duden
Grammar (1995) distinguishes between loosely constructed appositions (lockere Apposition) (e.g. Ute Wedemeier, die Landesvorsitzende,), which follow the head noun separated
by a comma, and tightly constructed appositions (enge Apposition) (e.g. (die) Landesvorsitzende Ute Wedemeier), which precede the head noun (cf. (Drosdowski 1995)). According to Helbig/Buscha (1998) there is case agreement between loosely constructed appositions and head nouns which are separated by a punctuation mark. By contrast, Engel
(1996) thinks that only loosely constructed appositions can be regarded as appositions.
He treats tightly constructed appositions as nomen varians or nomen invarians.
Because of these different definitions of the notion of apposition, we do not decide
on what is the head noun and what is the apposition. We assume referential identity
between the two parts. Loosely constructed appositions as well as tightly constructed appositions are treated as appositional constructions, i.e., the head noun and its apposition
34
form a complex structure which does not give any information about head assignment.
Therefore, both parts are first projected to their phrase level and then coordinated on a
higher level, each of them labelled as apposition (APP), i.e. as a part of an appositional
structure. What is important is the referential identity in meaning. Thus, Nummer 1 is
an appositional construction, whereas Seite 1 is a noun phrase with the postmodifier 1.
Forms of address for persons and titles, e.g. Herr, Frau, Doktor (Dr.), Professor (Prof.),
Bundeskanzler, are also treated as appositional constructions. Here are some examples:
NX
505
APP
APP
NX
503
NX
504
HD
−
NX
500
−
−
ADVX
501
HD
ADJX
502
HD
Donnerstag
HD
HD
morgen
0
,
1
den
2
13.
3
Mai
4
5
NN
ADV
$,
ART
ADJA
NN
asm
−−
−−
asm
asm
asm
NX
502
APP
APP
NX
500
NX
501
HD
HD
Herr
Taake
0
1
NN
NE
nsm
nsm
NX
502
APP
APP
NX
NX
500
501
HD
−
Landesvorsitzende
−
Ute
0
Wedemeier
1
2
NN
NE
NE
nsf
nsf
nsf
35
NX
505
APP
APP
NX
504
HD
−
NX
503
−
NX
HD
ADJX
500
−
−
Volker
NX
501
502
HD
Tegeler
0
,
1
−
stellvertretender
2
Geschäftsführer
3
HD
des
4
Landesverbandes
5
6
NE
NE
$,
ADJA
NN
ART
NN
nsm
nsm
−−
nsm
nsm
gsm
gsm
NX
502
APP
APP
NX
500
−
NX
501
HD
die
HD
Stadt
0
Frankfurt
1
2
ART
NN
NE
nsf
nsf
nsn
NX
504
APP
APP
NX
503
APP
APP
NX
500
NX
501
NX
502
HD
HD
HD
Vorwurf
Nummer
0
1
1
2
NN
NN
CARD
nsm
nsf
−−
NX
502
APP
APP
NX
500
NX
501
HD
HD
Telefon
472711
0
1
NN
CARD
dsn
−−
In case of a form of address combined with one or more titles preceding a name, we
annotate an embedded appositional construction:
36
NX
505
APP
APP
NX
NX
503
−
504
−
HD
ADJX
APP
APP
NX
NX
500
501
HD
Die
502
HD
Dortmunder
Psychologin
0
−
Prof.
1
2
−
Alexa
3
Franke
4
5
ART
ADJA
NN
NN
NE
NE
nsf
***
nsf
nsf
nsf
nsf
The same way, we treat proper nouns which are identical to a preceding proper noun,
for example, an actor’s name and role:
NX
502
APP
APP
NX
NX
500
−
501
−
Andrea
−
Spatzek
/
0
1
−
Gabi
2
Zenker
3
4
NE
NE
$(
NE
NE
dsf
dsf
−−
dsf
dsf
Premodification of the whole appositional construction is attached to an additional
NX level.
NX
504
−
HD
NX
503
ADVX
500
APP
APP
NX
NX
501
HD
502
HD
Auch
−
Bundesumweltminister
0
−
Jürgen
1
Trittin
2
3
ADV
NN
NE
NE
−−
nsm
nsm
nsm
There are some examples in which the appositional construction does not agree in
case. These are postnominal titles of books, films, etc. and translations interspersed in
the sentence. In the latter type, we extend the appositional construction also to nonnominal phrases.
37
PX
503
−
HD
NX
502
APP
APP
NX
NX
500
501
−
in
HD
dem
0
−
Film
1
"
2
HD
Das
3
Verhör
4
"
5
6
APPR
ART
NN
$(
ART
NN
$(
d
dsm
dsm
−−
nsn
nsn
−−
SIMPX
510
−
−
−
MF
509
OA
OPP
PX
508
APP
NCX
505
−
−
C
500
−
um
APP
PX
506
HD
−
VC
507
HD
HD
ADJX
501
HD
NCX
502
HD
−
die
wildernden
(
"
outlaw
"
)
zu
stellen
ADJA
NN
APPR
NN
$(
$(
FM
$(
$(
PTKZU
VVINF
−−
apm
apm
apm
g
gsn
−−
−−
−−
−−
−−
−−
−−
4.2.5
2
3
4
Gesetzes
HD
ART
1
außer
VXINF
504
KOUI
0
Hunde
FX
503
HD
5
6
7
8
9
10
11
12
Foreign Language Material
Words or parts of a text written in a foreign language except foreign language proper
nouns are tagged as foreign language material (FM), e.g. hello (FM), no (FM) longer
(FM) amused (FM). All parts of foreign language proper nouns are tagged as NE (e.g.
Mary (NE) , New (NE) York (NE), University (NE) of (NE) Illinois (NE)). Single foreign
words are projected to a syntactic level assigned the node label FX, which is an universal
label for any syntactic category (phrasal and sentential) in the respective foreign language.
More complex parts of a text tagged as FM are attached on the same level without any
internal syntactic structure and head assignment. Their mother node is also assigned the
label FM, e.g. no longer amused. For foreign language constructions containing a foreign
language proper noun, the annotation strategy is the following: in a first step, all NEs are
projected to the phrase level (NX), in a second step, these phrase node labels together
with all FMs are projected to the next higher level with the node label FX. Again, there
is no head assignment directly below the FX node, e.g. Mister Gere himself.
38
DM
500
−
hello
0
FM
−−
NX
501
−
−
FX
500
−
−
das
no
0
HD
longer
1
−
2
amused
3
Kollegium
ART
FM
FM
FM
NN
nsn
−−
−−
−−
nsn
4
NX
500
−
−
University
−
of
0
Illinois
1
2
NE
NE
NE
dsf
dsf
dsn
NX
502
−
HD
FX
501
−
−
−
NX
500
HD
wie
Mr.
0
Gere
1
himself
2
3
KOKOM
FM
NE
FM
−−
−−
nsm
−−
Often, foreign language material is a part of a German syntactic construction and plays
the role of a grammatical function. Therefore, the FX node is attached as a constituent
to the tree structure. If it is directly attached to a field or a sentence bracket, the edge
label above the FX node denotes its grammatical function within the clause, e.g. Kafka
goes Kleinkunst (head of the clause).
39
SIMPX
506
−
−
−
VF
503
LK
504
ON
HD
NX
500
FX
501
NX
502
HD
HD
HD
Kafka
MF
505
V−MOD
goes
Kleinkunst
0
1
2
NE
FM
NN
nsm
−−
asf
If a single FM is head of a phrase which can be identified as a German phrase, e.g.
by an article and/or an adjective (noun phrase), it is projected to the specific phrasal
category, e.g. NX instead of FX in a construction like in der Creme de la Kunst, die
nordamerikanischen Brothers.
PX
503
−
HD
NX
502
−
HD
FX
501
−
−
−
−
NX
500
HD
in
der
0
Creme
1
de
2
la
3
Kunst
4
5
APPR
ART
FM
FM
FM
NN
d
dsf
−−
−−
−−
dsf
NX
501
−
−
HD
ADJX
500
HD
ihrer
nordamerikanischen
0
Brothers
1
2
PPOSAT
ADJA
FM
gp*
gp*
−−
If FX is modified by a postmodifier the mother node of the complex phrase is also
FX, which again may be preceded by another phrase, e.g. Unter der Überschrift ’user als
looser’..
40
PX
505
−
HD
NX
504
APP
APP
FX
503
HD
NX
500
−
Unter
−
FX
501
HD
HD
NX
502
−
der
Überschrift
’
user
APPR
ART
NN
$(
FM
KOKOM
FM
$(
d
dsf
dsf
−−
−−
−−
−−
−−
0
4.2.6
1
2
3
4
als
HD
5
looser
6
’
7
Named Entity Annotation
Proper nouns denote individual living beings, objects, titles, etc. which exist only once
as entities with their own specific properties. The distinction between proper nouns and
nouns is not always clear-cut. On the one hand, proper nouns can also become nouns,
e.g. Opel as the company is a proper noun POS-tagged as Opel (NE), on the other hand,
Opel as the car is a noun POS-tagged as Opel (NN).
In addition to the categories of proper nouns listed in the STTS guidelines (e.g. first
and last names of persons, names of companies, and geographical names), we also define
names of streets and places (e.g. Feldstraße), individual names of institutions (e.g. MaxPlanck-Institut, Pergamonmuseum), events (e.g. Märzrevolution), and titles of books,
movies, etc. as specific categories of proper nouns . In contrast to the categories defined
in the STTS, our additional categories are not POS-tagged as ‘NE’, e.g. Alexanderplatz
(NN), heute (ADV).
Complex proper nouns forming a syntagma as well as titles, names of historical events,
institutions, and so on, are POS-tagged according to their distribution (e.g. der (ART)
Potsdamer (ADJA) Platz (NN), Auf (APPR) die (ART) stürmische (ADJA) Art (NN),
Schlaflos (ADJD) in (APPR) Seattle (NE)). On the syntactic level, we define all kinds of
proper nouns as named entities.
In our enhanced version of the TüBa-DZ treebank named entity information is defined
by semantic classes: each named entity is assigned one of the following five semantic
classes (cf. 3.4.2):
1. person (PER)
2. organisation (ORG)
3. location (LOC)
4. geopolitical entity (GPE)
5. other (OTH)
41
Table 4.1 lists the commonly occurring semantic subclasses for named entities in the
TüBa-D/Z with examples:
Table 4.1: Semantic Classes and Subclasses for Named Entities
Semantic Common Semantic Subclasses
Classes
PER
persons
surnames
names of animals (personified)
ORG
organisations
companies
institutes
museums
newspapers, journals
clubs
theaters, cinemas
universities
TV and radio stations
restaurants, hotels
forces
fashion labels
sporting events
bands
LOC
districts
sights, churches
planets
geographical areas
streets, places
mountains, lakes
continents
GPE
countries, states (incl. historical)
cities (incl. historical)
OTH
operating systems
titles of books, movies, etc.
mottoes, slogans
wars
Examples
Hans Winkler
(Familie) Feuerstein
(Schweinchen) Babe
Nato, EU
Microsoft, Bertelsmann,
Institut für chinesische Medizin
Pergamonmuseum
Süddeutsche Zeitung, Der Spiegel
VfB Stuttgart
Metropol-Theater, CinemaxX
Freie Universität
Arte, Radio Bremen
Sassella, Adlon
Blauhelme
Chanel
Olympische Spiele, Wimbledon
Beatles, Die Fantastischen Vier
Schöneberg
Brandenburger Tor, Johanniskirche
Mars
Königsheide
Sögestraße, Alexanderplatz
Alpen, Viktoriasee
Europa, Asien
Frankreich, Hessen, Assyrien
Berlin, Babylon
DOS
Faust, Schlaflos in Seattle
Zwischen Himmel und Erde
Zweiter Weltkrieg
In order to annotate the semantic classes, syntactic-semantic node labels of the
pattern syntactic category = semantic class are defined for the mother node of named
entities (see Table 3.9). The syntactic-semantic nodes indicate that the structure below
represents a (complex) named entity of a certain syntactic category belonging to one of
the five semantic classes (cf. 3.4.2).
The former node label ’EN-ADD’ and the secondary edge label ’EN’ are deleted.
42
Our annotation strategy for named entities is shown in the following tree examples:
Named entities may consist of one or more lexical elements tagged as NE. In case of
a single NE, this NE is projected to its phrase level, carrying the respective syntacticsemantic node label (Gütersloh (NX=GPE)). Named entities consisting of two or more
NEs are attached on the same level. None of them carries a head label in order to indicate
that there is no obvious dependency relation between them (St. Pauli (NX=LOC)).
There can occur postmodifiers within a named entity which are a named entity themselves like St. Pauli (NX=LOC) in [[FC (NX)] [St. Pauli (NX=LOC)]] (NX=ORG).
Parts which do not belong to a named entity are marked with the edge label -NE as
in [[den (-NE) FC (NX)] [Güthersloh (NX=GPE)]] (NX=ORG).
Named entities which are not tagged as NE, e.g. Millerntor (NX=LOC) are also
assigned a syntactic-semantic node label.
SIMPX
517
−
−
−
MF
515
V−MOD
V−MOD
VF
PX
514
512
ON
−
NX=ORG
−
NX
NX=LOC
500
VXFIN
−
St.
NX
502
HD
Pauli
1
503
−NE
1:0
3
gegen
4
−
NX
NX=GPE
−
HD
NX=LOC
505
HD
den
5
513
HD
504
HD
siegt
2
PX
509
HD
501
−
0
NX=ORG
508
HD
FC
HD
LK
511
HD
V−MOD
FC
6
506
HD
HD
Gütersloh
7
am
8
Millerntor
9
.
10
11
NN
NE
NE
VVFIN
CARD
APPR
ART
NN
NE
APPRART
NN
$.
nsm
nsm
nsm
3sis
−−
a
asm
asm
asn
dsn
dsn
−−
As mentioned above, all elements of German named entities which consist of a complex syntactic structure, e.g. a phrase or a sentence, are always tagged according to their
distribution and annotated with their internal syntactic structure as noun phrases, prepositional phrases, adjectival phrases, clauses, etc., e.g. the movie title (Schlaflos (ADJD)
in (APPR) Seattle (NE)) (ADJX=OTH) in the following tree example. If two named
entity nodes are coordinated like Tom Hanks (NX=PER) and Meg Ryan (NX=PER)
their mother node is NX which represents the nominal status of the named entity.
43
SIMPX
515
−
−
−
VF
514
V−MOD
PX
513
−
HD
ADJX=OTH
MF
511
512
HD
−
ON
PX
LK
507
508
−
ADJX
"
KONJ
PX=GPE
VXFIN
NX=PER
NX=PER
503
504
501
502
HD
Schlaflos
in
3
510
HD
HD
Seit
NX
509
HD
500
"
PRED
NX
HD
Seattle
4
"
5
−
gelten
6
KONJ
−
Tom
7
−
−
Hanks
8
und
9
HD
−
NX
NX
505
−
Meg
10
−
506
HD
Ryan
11
als
12
−
Dream−Team
13
HD
des
14
Biedersinns
15
.
0
1
2
$(
APPR
$(
ADJD
APPR
NE
$(
VVFIN
NE
NE
KON
NE
NE
KOKOM
NN
ART
NN
16
$.
−−
d
−−
−−
d
dsn
−−
3pis
nsm
nsm
−−
nsf
nsf
−−
nsn
gsm
gsm
−−
17
If the original form of a named entity (e.g. Zweiter Weltkrieg) is inflected and/or
premodified by an article and/or attributive adjective like in the following example tree
(dem Zweiten Weltkrieg (NX=OTH)), the mother node of the named entity carries the
semantic class information and all parts that do not belong to the named entity are
assigned the edge label -NE.
SIMPX
513
−
−
−
−
MF
512
MOD
V−MOD
PX
511
−
DM
VF
506
LK
507
−
VXFIN
500
HD
NX
VXFIN
HD
Stimmt
,
0
1
NX=OTH
508
ON
501
HD
HD
−NE
OV
VXINF
504
HD
505
HD
erst
3
HD
ADJX
503
wurde
2
510
−
ADVX
502
HD
sie
VC
509
nach
4
dem
5
HD
Zweiten
6
Weltkrieg
7
gebaut
8
.
9
10
VVFIN
$,
PPER
VAFIN
ADV
APPR
ART
ADJA
NN
VVPP
$.
3sis
−−
nsf3
3sit
−−
d
dsm
dsm
dsm
−−
−−
If a named entity of a syntactic category other than NX has a premodifier, or a
postmodifier, (both can be a named entity itself) the mother node of the whole constituent
is always NX which represents the nominal status of the named entity:
44
SIMPX
512
−
−
−
VF
MF
514
511
ON
V−MOD
NX
PX
513
−
510
−
HD
−
PX=OTH
−
ADJX
500
501
−
−
Oliver
Bukowskis
0
derbes
"
2
Bis
3
508
HD
HD
−
NX=GPE
VXFIN
ADJX
503
HD
1
NX
507
502
HD
HD
LK
506
NX=PER
OA
"
5
NX
504
HD
Denver
4
HD
505
HD
feierte
6
im
7
HD
Altonaer
8
Theater
9
Premiere
10
11
NE
NE
ADJA
$(
APPR
NE
$(
VVFIN
APPRART
ADJA
NN
NN
gsm
gsm
nsn
−−
a
asn
−−
3sit
dsn
***
dsn
asf
NX
508
HD
−
SIMPX=OT
507
−
−
LK
MF
504
506
HD
ON
PRED
VXFIN
NX
NX=PER
500
HD
"
PX
505
501
HD
Sind
−
HD
NX=PER
502
503
HD
Sie
−
Luigi
?
3
"
von
−
Stephan
Brüggenthies
0
1
2
4
5
6
$(
VAFIN
PPER
NE
$.
$(
APPR
NE
7
NE
8
−−
3pis
np*3
nsm
−−
−−
d
dsm
dsm
If a named entity is a syntagma with its own internal syntactic structure, i.e. it does
not agree with the inflection of another constituent of the sentence, all premodifiers are
attached on a higher level:
45
SIMPX
511
−
−
−
MF
510
ON
VF
NX
507
509
OPP
−
PX
HD
LK
504
NX=OTH
505
−
HD
HD
NX
VXFIN
500
NX=PER
501
HD
Im
506
−
Radio
1
503
HD
läuft
0
ADJX
502
HD
HD
Vivaldis
2
HD
"
3
Vier
Jahreszeiten
4
5
"
6
.
7
8
APPRART
NN
VVFIN
NE
$(
CARD
NN
$(
$.
dsn
dsn
3sis
gsm
−−
−−
npf
−−
−−
SIMPX
512
−
−
−
MF
511
ON
NX
510
HD
−
PX
509
−
HD
NX
508
−
VF
505
507
HD
NX
−
VXFIN
500
NX
501
HD
Den
PX=OTH
506
OA
−
HD
LK
Auftakt
0
−
macht
1
NX=PER
502
HD
NX
503
HD
eine
2
HD
−
Aufführung
3
von
4
Ernst
5
504
−
−
Jandls
6
Aus
7
HD
der
8
Fremde
9
.
10
11
ART
NN
VVFIN
ART
NN
APPR
NE
NE
APPR
ART
NN
$.
asm
asm
3sis
nsf
nsf
d
gsm
gsm
d
dsf
dsf
−−
If a named entity has a postmodifier that is part of the named entity itself, all premodifiers are also attached on a higher level:
46
NX
507
HD
−
NX
506
−
−
HD
NX
NX=OTH
504
−
505
−
HD
ADJX
ADJX
500
501
HD
Das
167seitige
Aktionsprogramm
1
der
2
NX
NX
503
HD
Bremer
3
−
502
HD
0
HD
HD
Agenda
4
21
5
6
ART
ADJA
NN
ART
ADJA
NN
CARD
nsn
nsn
nsn
gsf
***
gsf
−−
Foreign Language Named Entities
The syntactic annotation of foreign language named entities differs from the annotation
of German named entities in the following aspects. According to the STTS guidelines,
foreign language proper nouns are tagged as NE, while all other lexical elements of a
foreign language are tagged as foreign language material (FM). A foreign language named
entity which consists of a proper noun, e.g. the title of a movie is assigned a syntacticsemantic node label of the category NX (Forrest Gump (NX=OTH)).
NX=OTH
500
−
−
Forrest
Gump
0
NE
NE
nsm
nsm
1
If a foreign language named entity consists of only FM tagged tokens, these tokens
are directly attached on the same level without internal syntactic structure. The mother
node of the phrase is marked as a syntactic-semantic node label of the category FX, e.g.
Knockin’ on Heaven’s Door (FX=OTH).
FX=OTH
500
−
Knockin’
−
0
on
−
1
Heaven’s
−
2
Door
FM
FM
FM
FM
−−
−−
−−
−−
3
If a foreign language named entity consists of NE as well as FM tagged tokens, e.g.
Shakespeare (NE) in (FM) Love (FM), NE is projected to NX=PER. The NX=PER node
and all FM tagged tokens are attached directly on the same level.
47
FX=OTH
501
−
−
−
NX=PER
500
HD
Shakespeare
in
Love
NE
FM
FM
nsm
−−
−−
4.2.7
0
1
2
Ordinal Numbers
According to their distribution, ordinal numbers occur either as a premodifying attributive adjective (e.g. die dritte (ADJA) Partie) or as a head noun (e.g. er ist der sechste
(NN)). In the first case, the premodifier is projected to an adjectival phrase, in the latter
case it is projected to a noun phrase.
NX
501
−
−
HD
ADJX
500
HD
die
dritte
0
Partie
1
2
ART
ADJA
NN
nsf
nsf
nsf
SIMPX
510
−
−
KOORD
500
−
VF
507
LK
508
ON
HD
MOD
MOD
MOD
VXFIN
502
ADVX
503
ADVX
504
ADVX
505
HD
HD
HD
HD
ja
auch
NX
501
−
HD
Aber
−
das
0
ist
1
2
MF
509
3
PRED
NX
506
−
schon
4
HD
die
5
Vierte
6
.
7
8
KON
PDS
VAFIN
ADV
ADV
ADV
ART
NN
$.
−−
nsn
3sis
−−
−−
−−
nsf
nsf
−−
4.2.8
Cardinal Numbers
According to their syntactic function (nominal or adjectival), cardinal numbers (CARD),
are either projected to NX or ADJX. If their numerals are written separately or in groups,
e.g. numbers of bank accounts, they are attached on the same level like proper names
without internal head assignment.
48
NX
502
APP
APP
NX
500
NX
501
HD
HD
Jahr
2000
0
1
NN
CARD
dsn
−−
PX
502
−
HD
NX
501
−
−
HD
ADJX
500
HD
in
allen
0
23
1
Bezirken
2
3
APPR
PIDAT
CARD
NN
d
dpm
−−
dpm
NX
502
APP
APP
NX
500
NX
501
HD
−
BLZ
−
500
0
−
901
00
1
2
NN
CARD
CARD
CARD
3
nsf
−−
−−
−−
A premodifying cardinal number is nominal if it does not express a quantity like in
the example above, but a characteristic of the following noun, e.g. the number of a zip
code:
NX
501
−
HD
NX
500
HD
13187
Berlin
0
1
CARD
NE
−−
nsn
Complex time expressions or results of competitions are also treated as cardinal numbers:
49
NX
501
−
HD
ADJX
500
HD
20.15
Uhr
0
1
CARD
NN
−−
nsf
PX
501
−
HD
NX
500
HD
mit
3:0
0
1
APPR
CARD
d
−−
4.2.9
Letters and Non-Words
Letters and non-words are tagged as XY. They are projected to their phrase level and
assigned the syntactic category to which they belong in the construction. Signs which
represent a lexical element, e.g. the sign for paragraph, are tagged with the respective
part-of-speech tag:
NX
501
−
HD
NX
500
HD
D−76351
Linkenheim
0
1
XY
NE
−−
dsn
NX
506
HD
−
NX
505
HD
−
NX
504
HD
−
NX
500
NX
501
NX
502
NX
503
HD
HD
HD
−
§
220
a
des
HD
Strafgesetzbuches
0
1
2
3
NN
CARD
XY
ART
NN
4
nsm
−−
−−
gsn
gsn
50
4.2.10
Expletive and Other Uses of es
The pronominal form es functions as expletive element in German. Three different expletive usages are traditionally distinguished: formal subject or object, correlate of an
extraposed clausal argument, and Vorfeld-es (cf. (Eisenberg 1999 2001), (Pütz 1986)).
For sake of completeness, the following list begins with an example of es as a referential
personal pronoun.
Personal Pronoun
The pronoun functions as an argument of the verb and refers to some person, object, or
event that is salient in the context. It can be tested, whether es is used as a pronoun by
replacing it by another noun or pronoun (such as das or er/ihn).
In the example tree es refers to the neuter noun Gästehaus in the preceding sentence:
Die italienische Regierung hat die Familie im staatlichen Gästehaus Casino dell’Algardi
untergebracht.
SIMPX
509
−
−
−
−
MF
508
FOPP
VF
504
LK
505
ON
HD
NX
500
HD
PX
506
−
HD
VXFIN
501
VXINF
503
HD
wird
0
OV
NX
502
HD
Es
VC
507
von
HD
Scharfschützen
bewacht
3
.
1
2
4
PPER
VAFIN
APPR
NN
VVPP
$.
5
nsn3
3sis
d
dpm
−−
−−
Formal Subject or Object
The formal subject obligatorily occurs with weather verbs, e.g. Es regnet and impersonal
or agentless constructions such as Es gibt so eine Buchung or Es geht um populäre Unterhaltung. Some verbs optionally permit an expletive subject but also occur with referential
subjects such as Max/Es klopft an der Tür. A formal object is found in constructions
like jmd. legt es an auf etw. or jmd. verdirbt es mit jmdm. In all examples mentioned,
es functions as a grammatical argument without semantic contribution, i.e. it does not
refer to a person, object, or event.
In TüBa-D/Z formal subjects and objects are treated like referential pronouns and
are labelled alike, e.g. with edge labels ON or OA.
Formal arguments are obligatory and may occur in the Mittelfeld. In case of doubt, it
is a good test to paraphrase the sentence such that another element occupies the Vorfeld,
e.g. Natürlich gibt es so eine Buchung versus *Natürlich gibt so eine Buchung.
51
SIMPX
507
−
−
−
MF
506
OA
VF
503
LK
504
ON
HD
−
VXFIN
501
ADVX
502
HD
HD
NX
500
HD
"
Es
0
gibt
1
NX
505
−
so
HD
eine
Buchung
4
.
5
"
2
3
6
7
$(
PPER
VVFIN
ADV
ART
NN
$.
$(
−−
nsn3
3sis
−−
asf
asf
−−
−−
Correlate of an Extraposed Clausal Argument
If a clausal argument is extraposed in the Nachfeld, it is optionally doubled by an expletive
es in the Vorfeld or Mittelfeld. The expletive is labelled ON-MOD or OS-MOD depending
on the function of the clausal argument.
−
−
SIMPX
521
−
−
−
NF
520
ON
SIMPX
519
−
−
NF
518
OS
SIMPX
517
−
VF
510
ON−MOD
KOORD
500
−
Aber
NX
501
HD
−
LK
511
HD
MF
512
PRED
VC
513
HD
VF
514
V−MOD
LK
515
HD
VXFIN
502
HD
ADJX
503
HD
VXINF
504
PX
505
HD
VXFIN
506
HD
HD
−
ON
NX
507
−
ADVX
508
HD
HD
NX
509
−
es
ist
übertrieben
zu
sagen
,
damit
die
FU
VAFIN
ADJD
PTKZU
VVINF
$,
PROP
VVFIN
ART
NE
ADV
ART
NN
$.
−−
nsn3
3sis
−−
−−
−−
−−
−−
3skt
nsf
nsf
−−
asf
asf
−−
2
3
4
5
6
7
8
9
10
11
eine
HD
PPER
1
erst
OA
KON
0
bekäme
−
MF
516
MOD
12
Identität
13
Vorfeld-es
The last type is a purely structural dummy element. It occurs in Vorfeld position only
and is not correlated with any argument of the clause. It does not agree with the verb
which becomes evident if there is a plural subject in the Mittelfeld, which is illustrated
in the example tree below. It is ungrammatical in the Mittelfeld, e.g. *. . . dass es ihn die
Völker zahlen. Vorfeld-es is labelled ES to indicate its purely structural function. In the
first release of TüBa-D/Z, 12/2003, Vorfeld-es was integrated by means of ON-MOD.
52
.
14
SIMPX
516
−
−
−
−
NF
515
ON−MOD
R−SIMPX
514
−
VF
508
ES
LK
509
MF
510
HD
NX
500
HD
VXFIN
501
HD
es
ON
ON
MOD
NX
502
NX
503
NX
504
ADJX
505
zahlen
−
ihn
1
HD
die
2
−
Völker
3
,
4
HD
deren
−
MF
512
OA
HD
0
−
C
511
VC
513
OV
HD
VXINF
506
VXFIN
507
HD
HD
HD
Menschenrechte
angeblich
7
verteidigt
8
werden
5
6
PPER
VVFIN
PPER
ART
NN
$,
PRELAT
NN
ADJD
VVPP
9
VAFIN
10
****
3pis
asm3
npn
npn
−−
gp*
npn
−−
−−
3pis
11
Table 4.2 summarizes tests and labels for the different uses of es.
Table 4.2: Types of es
type
test
substitutable
by other pronouns
optional
referential
pronoun
yes
formal
argument
no
correlate
no
Vorfeld-es
no
no
no
yes
no
correlates with
clausal argument
ungrammatical
in Mittelfeld
edge label
no
no
yes
no
no
no
no
yes
ON, OA,
etc.
ON, OA
ON-MOD,
OS-MOD
ES
Es sei denn
The lexicalized phrase es sei denn, meaning außer, is analysed as a copula construction.
53
12
SIMPX
523
KONJ
KONJ
SIMPX
522
−
−
−
−
NF
521
PRED
SIMPX
519
−
−
VF
510
−
LK
511
ES
NX
500
HD
"
SIMPX
520
Es
HD
V−MOD
VXFIN
501
ADVX
502
HD
HD
geschieht
0
1
−
MF
512
hier
ON
ON
NX
503
NX
504
HD
HD
nichts
3
,
4
es
5
LK
514
−
VF
516
HD
MOD
ON
VXFIN
505
ADVX
506
NX
507
HD
HD
sei
6
MF
515
HD
denn
7
,
8
−
LK
517
MF
518
HD
OA
VXFIN
508
HD
ich
NX
509
HD
tu
es
11
.
12
13
"
9
10
$(
PPER
VVFIN
ADV
PIS
$,
PPER
VAFIN
ADV
$,
PPER
VVFIN
PPER
$.
$(
−−
****
3sis
−−
***
−−
nsn3
3sks
−−
−−
ns*1
1sis
asn3
−−
−−
4.3
2
VF
513
Determiner Phrases
Certain pronouns serving as determiners in noun phrases may be premodified, for instance, by degree adverbs such as in so viele Ältere, gar kein Schutz, etc.
In the case of so viele Ältere, the premodifying adverb so is attached to the indefinite
pronoun viele. Together, they form a determiner phrase (DP), which is attached to the
head noun Ältere on the same level:
NX
502
−
HD
DP
501
−
HD
ADVX
500
HD
so
viele
Ältere
0
1
ADV
PIDAT
NN
2
−−
ap*
ap*
NX
502
−
HD
DP
501
−
HD
ADVX
500
HD
gar
kein
0
Schutz
1
2
ADV
PIAT
NN
−−
nsm
nsm
54
14
4.4
Prepositional Phrases
4.4.1
Prepositions
Considering prepositional phrases, it turns out to be appropriate not to annotate the
preposition as the head of the phrase. It is rather reasonable to annotate the complement
within the prepositional phrase as the head. This decision facilitates the identification of
dependencies between verbs and their nominal complements and adjuncts. Moreover, it
is in accordance with basic assumptions in Dependency Grammar .
PX
501
−
HD
NX
500
HD
in
Südpolen
APPR
NN
d
dsn
0
1
If the preposition is realized as a non-alphabetic sign, e.g. - (bis, gegen), this sign is
tagged as APPR and annotated like a preposition:
NX
504
HD
−
PX
503
−
HD
NX
502
−
NX
ADJX
500
−
501
−
HSV
HD
HD
BU
0
−
1
Bramfelder
2
SV
3
4
NE
NE
APPR
ADJA
NN
nsm
nsn
a
***
asm
Since pronominal adverbs (PROP) are pronominal forms of a prepositional phrase,
they are directly projected to PX:
55
SIMPX
510
−
−
−
−
VF
506
ON
LK
507
HD
V−MOD
MF
508
OA
NX
500
HD
VXFIN
501
HD
ADVX
502
HD
NX
503
HD
Freuden−thal
0
wollte
1
gestern
2
nichts
VC
509
OV
FOPP
3
PX
504
HD
VXINF
505
HD
dazu
sagen
4
NE
VMFIN
ADV
PIS
PROP
VVINF
nsf
3sit
−−
***
−−
−−
5
Freudenthal
In German, there are so-called Verschmelzungsformen, i.e. merged forms of a preposition and a determiner, e.g. in dem Januar amalgamates to im Januar. The merged form is
assigned the part-of-speech tag APPRART (including richer morphological annotation).
In terms of syntax, it is annotated like a preposition:
PX
501
−
HD
NX
500
HD
Im
Januar
0
1
APPRART
NN
dsm
dsm
Prepositional phrases expressing intervals, e.g. with von/bis, von/bis zu or zwischen,
are annotated in the same way as coordinate structures (cf. 6.4.1), i.e. without head
assignment on the level of coordination, since the two phrases are assumed to be conjuncts.
If two prepositions follow each other (e.g. bis zum), the result is an embedded structure
of a prepositional phrase taking another preposition. The first preposition does hereby
not receive a morphological case feature.
PX
506
KONJ
KONJ
PX
504
−
PX
505
HD
−
HD
NX
502
NX
503
−
−
ADJX
500
ADJX
501
HD
vom
HD
23.
0
HD
bis
1
25.
2
Juli
3
4
APPRART
ADJA
APPR
ADJA
NN
dsm
dsm
a
asm
asm
56
PX
505
−
HD
NX
504
KONJ
−
KONJ
NX
502
NX
503
−
−
ADJX
500
ADJX
501
HD
zwischen
HD
15.000
0
und
1
22.000
2
3
APPR
CARD
KON
CARD
d
−−
−−
−−
PX
504
−
HD
PX
503
−
HD
NX
502
APP
bis
zum
0
APP
NX
500
NX
501
HD
HD
Jahr
1
2000
2
3
APPR
APPRART
NN
CARD
−−
dsn
dsn
−−
As opposed to the case with two prepositions, intervals like dritter bis fünfter November are annotated as a coordinate attributive adjective phrase within a simple noun phrase
(cf. 6.4.1).
Premodification of non-isolated prepositional phrases follows the general principle of
low attachment.
PX
504
−
−
HD
NX
503
HD
ADVX
500
HD
Irgendwo
−
NX
501
−
NX
502
HD
HD
in
den
ADV
APPR
ART
NN
NE
−−
d
dpm
dpm
gsn
0
1
2
Wäldern
3
Schaumburgs
4
There is one exception to the low attachment principle: isolated phrases in which a
preceding adverb does not semantically modify the prepositional phrase. In this case the
adverbial phrase is high attached to an additional level of PX.
57
PX
503
−
HD
PX
502
−
HD
ADVX
500
HD
Nun
NX
501
HD
zum
0
Wetter
1
ADV
APPRART
NN
−−
dsn
dsn
4.4.2
2
Circumpositions and Postpositions
Circumpositions are treated as ternary branching prepositional phrases. The circumposition on the left hand side is tagged as APPR and the circumposition on the right hand
side as APZR:
PX
501
−
HD
−
NX
500
HD
von
sich
0
aus
1
2
APPR
PRF
APZR
d
ds*3
−−
Postpositions are tagged as APPO. The complement of the postposition occurs on the
left side and constitutes the head of the prepositional phrase:
PX
501
HD
−
NX
500
−
HD
Dem
Vernehmen
0
nach
1
2
ART
NN
APPO
dsn
dsn
d
4.5
Adjectival Phrases
We distinguish between attributive adjectives on the one hand and adverbial or predicative adjectives respectively on the other hand. Attributive adjectives are tagged as
58
ADJA (die traditionellen Elemente) or CARD (20.15 Uhr), whereas adverbial or predicative adjectives are tagged as ADJD (das Gewicht ist gut; den betriebswirtschaftlich
günstigeren Standort) or PWAV (wie wirke ich).
The annotation of superlative and comparative forms is explained in section 7.1 on
page 112.
In general, German adjectives are inflected when they are an attribute of a noun. They
are not inflected either when they function as a predicative adjective or a premodifier of
an adjective or an adverb or when they belong to a small class of noninflected adjectives,
e.g. some ancient form such as gut Wetter or lieb Mütterlein or some adjectives denoting
a colour (mit einer rosa Karte). All adjectives have to be projected to their phrase level
before they are attached to another phrase or to a field.
NX
501
−
−
HD
ADJX
500
HD
Die
traditionellen
Elemente
0
1
2
ART
ADJA
NN
npn
npn
npn
PX
502
−
HD
NX
501
−
−
HD
ADJX
500
HD
mit
einer
0
rosa
1
Karte
2
3
APPR
ART
ADJA
NN
d
dsf
dsf
dsf
SIMPX
506
−
−
VF
503
LK
504
ON
HD
NX
500
−
HD
Gewicht
0
ADJX
502
HD
ist
1
MF
505
PRED
VXFIN
501
HD
Das
−
gut
2
3
ART
NN
VAFIN
ADJD
nsn
nsn
3sis
−−
59
NX
502
−
−
HD
ADJX
501
−
HD
ADJX
500
HD
den
betriebswirtschaftlich
0
1
günstigeren
2
Standort
ART
ADJD
ADJA
NN
asm
−−
asm
asm
SIMPX
509
−
−
−
3
−
VF
508
ON
NX
504
−
−
HD
ADJX
500
HD
Der
männliche
0
1
Trinker
2
LK
505
HD
MF
506
V−MOD
VC
507
OV
VXFIN
501
HD
ADJX
502
HD
VXINF
503
HD
sei
3
gut
4
erforscht
5
,
6
ART
ADJA
NN
VAFIN
ADJD
VVPP
$,
nsm
nsm
nsm
3sks
−−
−−
−−
A nominalized adjective like Fassbares might be premodified by an adverbial adjective
(ADJD) instead of an attributive adjective (ADJA). The former ones do never inflect.
NX
501
−
HD
ADJX
500
HD
physisch
Faßbares
0
1
ADJD
NN
−−
asn
Whenever an adjective is modified by another modifier, the same annotation strategy
as for noun phrases is applied, i.e., the modifier is directly attached to the adjectival
phrase. The adjectival phrase as a whole is the premodifier of the noun phrase. For
instance:
60
NX
502
−
−
HD
ADJX
501
−
HD
ADVX
500
HD
eine
sehr
0
gute
1
Quelle
2
3
ART
ADV
ADJA
NN
nsf
−−
nsf
nsf
SIMPX
508
−
−
−
−
MF
507
PRED
KOORD
500
VF
504
LK
505
ON
HD
−
VXFIN
502
ADVX
503
HD
HD
NX
501
−
−
aber
HD
der
0
Text
1
ADJX
506
ist
2
HD
sehr
3
abstrakt
4
5
KON
ART
NN
VAFIN
ADV
ADJD
−−
nsm
nsm
3sis
−−
−−
The same holds if an adjective selects an argument. Für die Weltgesellschaft is the
facultative argument of wesentlich. It is directly attached to the adjectival phrase.
NX
503
−
−
HD
ADJX
502
−
HD
PX
501
−
HD
NX
500
−
Die
für
0
HD
die
1
Weltgesellschaft
2
wesentliche
3
Unterscheidung
4
5
ART
APPR
ART
NN
ADJA
NN
nsf
a
asf
asf
nsf
nsf
Premodifying adjectives may occur in a linear order and/or as a coordination (cf.
6.4.1) of attributive adjectives:
61
NX
503
−
HD
ADJX
502
KONJ
−
KONJ
ADJX
500
ADJX
501
HD
HD
28.
und
0
29.
1
Mai
2
3
ADJA
KON
ADJA
NN
nsm
−−
nsm
nsm
NX
504
−
−
−
HD
ADJX
503
KONJ
KONJ
ADJX
500
ADJX
501
HD
Die
ADJX
502
HD
großen
0
,
1
HD
bekannten
serbischen
2
3
Oppositionsparteien
4
5
ART
ADJA
$,
ADJA
ADJA
NN
npf
npf
−−
npf
npf
npf
NX
504
−
−
−
HD
ADJX
503
KONJ
ADJX
500
ADJX
502
HD
eigene
0
KONJ
ADJX
501
HD
ihre
−
HD
demokratische
1
und
2
freiheitliche
3
Tradition
4
5
PPOSAT
ADJA
ADJA
KON
ADJA
NN
asf
asf
asf
−−
asf
asf
If the premodifying adjective is deverbal, the adjectival phrase can be of any complexity. In this case, the adjectival phrase has its own internal dependency structure.
All elements which depend on the adjective are annotated as its premodifiers. Deverbal
adjectives are either attributive or adverbial and predicative respectively, and occur as
the present participle or past participle form of a verb.
62
NX
502
−
−
HD
ADJX
501
−
HD
ADJX
500
HD
das
aktuell
0
diskutierte
Thema
1
2
3
ART
ADJD
ADJA
NN
asn
−−
asn
asn
NX
504
−
−
HD
ADJX
503
−
−
HD
PX
502
−
HD
ADVX
500
HD
Die
0
teilweise
NX
501
−
1
HD
in
die
Erde
2
3
4
gebaute
5
Sporthalle
ART
ADV
APPR
ART
NN
ADJA
NN
nsf
−−
a
asf
asf
nsf
nsf
6
In the following example, postmodification of an adjectival phrase is shown:
ADJX
502
HD
−
ADJX
500
ADJX
501
HD
−
besser
HD
als
0
gut
1
2
ADJD
KOKOM
ADJD
−−
−−
−−
4.6
Adverbial Phrases
Besides adverbials also negation particles (PTKNEG) project to an adverbial phrase.
They either occur as premodifiers1 or postmodifiers or they are directly attached to a
field.
1
bis zu, über are considered to be ADV rather than APPR because of their semantic meaning.
63
SIMPX
515
−
−
−
−
−
MF
514
OD
VF
LK
509
KOORD
ADJX
APP
ADVX
VXFIN
NX
NX
502
HD
503
HD
heute
0
will
−
Ina
1
2
−
NX
504
−
−
Terre
3
(
4
VC
512
APP
−
V−MOD
511
HD
501
MOD
NX
510
V−MOD
500
Doch
ON
−
Hannelore
5
ADVX
505
HD
Droege
6
)
7
8
VXINF
507
508
HD
nicht
9
OV
ADVX
506
HD
es
513
HD
HD
so
10
recht
11
munden
12
13
KON
ADV
VMFIN
NE
NE
$(
NE
NE
$(
PPER
PTKNEG
ADV
ADJD
VVINF
−−
−−
3sis
dsf
dsf
−−
dsf
dsf
−−
nsn3
−−
−−
−−
−−
NX
503
−
−
ADVX
500
ADVX
501
HD
HD
bis
−
HD
ADJX
502
HD
zu
300.000
Leute
0
1
2
ADV
ADV
CARD
NN
3
−−
−−
−−
np*
NX
502
−
−
ADVX
500
HD
ADJX
501
HD
HD
über
350.000
0
Auskünfte
1
2
ADV
CARD
NN
−−
−−
apf
ADVX
502
HD
−
ADVX
500
ADVX
501
HD
HD
heute
abend
0
1
ADV
ADV
−−
−−
64
SIMPX
510
−
−
−
−
MF
509
V−MOD
VF
505
LK
506
ON
HD
HD
−
OV
VXFIN
501
ADVX
502
ADVX
503
VXINF
504
HD
HD
HD
HD
NX
500
−
HD
Der
Fahrer
0
ADVX
507
konnte
1
nicht
2
VC
508
mehr
3
bremsen
4
.
5
6
ART
NN
VMFIN
PTKNEG
ADV
VVINF
$.
nsm
nsm
3sit
−−
−−
−−
−−
4.7
Verb Phrases
Whereas finite verb phrases are labelled VXFIN, non-finite verb phrases are labelled
VXINF.
Since infinitives and past participles share certain properties (e.g. exchangeability in
Man hat nur noch das eigene Herz schlagen hören/gehört.), they are assumed to carry
the same phrase label (VXINF). The finite verb in LK as well as the non-finite verbs in
VC are always projected to their phrase level. All verb phrases of the verb complex are
attached on the same level to form the verb complex. In order to follow the flat clustering
principle, no internal hierarchy of the verb complex is annotated.
4.7.1
Head of a Sentence and Verb Complex
The finite verb which can either appear in LK (verb-first clauses and verb-second clauses)
or in VC (verb-final clauses), is always the head of the entire sentence. Non-finite verbal
elements belong to VC. If the finite verb is located in LK and if there is more than one
non-finite element in VC, the non-finite element which is selected by the finite verb is
denoted as the head of VC. All other elements of VC are verbal objects. The head of VC
selects the verbal object OV. This verbal object may select another verbal object OV,
and so on. In order to denote the dependency relations between verbal objects within
the verb complex, we attach a secondary edge label refvc between their phrase nodes.
4.7.2
Verb Complexes in Verb-second and Verb-final Clauses
The following example shows a verb-second clause with the head of the sentence in LK
and a verb complex consisting of a single non-finite element.
65
SIMPX
513
−
−
−
−
MF
512
OA
FOPP
VF
PX
510
511
ON
−
NX
LK
506
−
NX
507
−
HD
NX
501
überehrgeizige
Bürgermeister
1
HD
die
3
−
Bergidylle
4
in
5
ein
VXINF
504
HD
6
OV
NX
503
−
will
2
−
NX
502
HD
0
509
HD
VXFIN
500
HD
VC
508
HD
ADJX
Der
HD
−
Mekka
7
des
8
505
HD
HD
Massentourismus
9
verwandeln
10
11
ART
ADJA
NN
VMFIN
ART
NN
APPR
ART
NE
ART
NN
VVINF
nsm
nsm
nsm
3sis
asf
asf
a
asn
asn
gsm
gsm
−−
If the verb complex comprises more than one immediate daughter, the one that is
selected by the finite verb is the head of VC.
SIMPX
509
−
−
−
VF
505
LK
506
ON
HD
NX
500
NX
502
HD
Es
VC
508
PRED
VXFIN
501
HD
−
MF
507
−
müsse
0
HD
ein
1
Buchungsfehler
2
OV
HD
VXINF
503
VXINF
504
HD
HD
gewesen
3
sein
4
.
5
6
PPER
VMFIN
ART
NN
VAPP
VAINF
$.
nsn3
3sks
nsm
nsm
−−
−−
−−
The following trees demonstrate verb complexes with two or more verbal objects. The
secondary edge label refvc is pointing from the selecting OV to the depending OV.
SIMPX
508
−
−
−
MF
506
MOD
VC
507
ON
OV
refvc
C
500
ADVX
501
−
HD
Wenn
HD
da
0
NX
502
was
1
OV
503
HD
VXINF
503
VXINF
504
VXFIN
505
HD
HD
HD
gebucht
2
worden
3
ist
4
5
KOUS
ADV
PIS
VVPP
VAPP
VAFIN
−−
−−
***
−−
−−
3sis
66
SIMPX
513
−
−
−
MF
512
MOD
ON
V−MOD
PX
511
−
HD
NX
509
HD
VC
510
−
OV
refvc
C
500
ADVX
501
−
NX
502
HD
daß
−
auch
0
HD
sein
1
Vater
2
auf
3
NX
503
NX
504
HD
HD
Anregung
4
Andreottis
5
OV
505
refvc
OV
506
HD
VXINF
505
VXINF
506
VXINF
507
VXFIN
508
HD
HD
HD
HD
umgebracht
6
worden
7
sein
8
könnte
9
10
KOUS
ADV
PPOSAT
NN
APPR
NN
NE
VVPP
VAPP
VAINF
VMFIN
−−
−−
nsm
nsm
a
asf
gsm
−−
−−
−−
3skt
If there is no finite verb at all, the rightmost element of the verb complex (if there is
more than one element) is annotated as the head of the sentence. This often occurs in
headlines (cf. 5.2 and 7.4).
SIMPX
504
−
−
MF
502
VC
503
OA
HD
NX
500
VXINF
501
HD
HD
Prachtwicken
gucken
0
.
1
2
NN
VVINF
$.
apf
−−
−−
4.7.3
Ersatzinfinitiv Constructions
In order to indicate Ersatzinfinitiv constructions, two specific field node labels are introduced. VCE is the node label for the part of the verb complex consisting of the finite
verb which subcategorizes for the Ersatzinfinitiv. MFE is the node label for the second
part of MF between VCE and the second part of the verb complex VC (e.g. [C die] [MF
uns] [VCE hätten] [MFE mißtrauisch] [VC machen müssen]).
67
SIMPX
511
−
−
−
−
MF
510
ON
OPP
NX
507
KONJ
VCE
508
−
KONJ
C
500
NX
501
NX
502
PX
503
−
HD
HD
HD
daß
Fischer
0
und
1
ich
2
dazu
3
VC
509
HD
OV
HD
VXFIN
504
VXINF
505
VXINF
506
HD
HD
HD
haben
4
beitragen
können
5
6
7
KOUS
NE
KON
PPER
PROP
VAFIN
VVINF
VMINF
−−
nsm
−−
ns*1
−−
1pis
−−
−−
R−SIMPX
511
−
−
−
−
−
C
506
ON
MF
507
OA
VCE
508
HD
MFE
509
PRED
OV
HD
NX
500
HD
NX
501
HD
VXFIN
502
HD
ADJX
503
HD
VXINF
504
HD
VXINF
505
HD
die
uns
PRELS
PPER
VAFIN
ADJD
VVINF
VMINF
np*
ap*1
3pkt
−−
−−
−−
0
hätten
VC
510
1
mißtrauisch
2
3
machen
4
müssen
5
In the example below, the finite verb precedes the non-finite verbs although müssen is
no Ersatzinfinitiv. Since its position corresponds to the position of the finite verb in real
Ersatzinfinitiv constructions and here also a second middle field is possible, we follow the
same annotation strategy.
SIMPX
514
−
−
−
−
MF
513
ON
OD
MOD
MOD
OA
NX
512
−
−
HD
ADJX
509
−
C
500
−
NX
501
−
daß
NX
502
HD
die
HD
Nato
sich
2
VCE
510
HD
OV
HD
ADVX
503
ADVX
504
ADVX
505
VXFIN
506
VXINF
507
VXINF
508
HD
HD
HD
HD
HD
HD
doch
3
noch
4
ein
5
HD
VC
511
ganz
6
neues
8
wird
9
überlegen
10
müssen
1
KOUS
ART
NE
PRF
ADV
ADV
ART
ADV
ADJA
NN
VAFIN
VVINF
VMINF
−−
nsf
nsf
ds*3
−−
−−
asn
−−
asn
asn
3sis
−−
−−
68
7
Konzept
0
11
12
4.7.4
Infinitives with zu
Regarding infinitives with zu, zu determines the non-finiteness of the verb on its right
hand side. This is the reason why zu is considered the head of the VXINF whereas the
infinitive is assumed to be the complement. Like other infinitives, they occur in the verb
complex:
SIMPX
512
−
−
−
−
VF
511
ON
NX
510
HD
−
NX
506
−
HD
NX
LK
501
KONJ
NX
−
NX
500
513
HD
der
0
Friedens−
und
2
509
HD
OD
PRED
OV
NX
VXFIN
NX
ADJX
VXINF
502
HD
1
VC
508
KONJ
514
−
Erkenntnisse
MF
507
503
HD
Konfliktforschung
3
HD
scheinen
4
505
HD
ihm
5
504
HD
fremd
6
−
zu
7
sein
8
.
9
10
NN
ART
TRUNC
KON
NN
VVFIN
PPER
ADJD
PTKZU
VAINF
$.
npf
gsf
−−
−−
gsf
3pis
dsm3
−−
−−
−−
−−
SIMPX
510
−
−
−
−
VF
509
OPP
PX
505
−
LK
506
HD
NX
500
HD
Über
Details
0
MF
507
VC
508
HD
MOD
OV
HD
VXFIN
501
ADVX
502
VXINF
503
VXINF
504
HD
HD
werde
1
HD
noch
2
−
zu
3
HD
verhandeln
4
sein
5
.
6
7
APPR
NN
VAFIN
ADV
PTKZU
VVINF
VAINF
$.
a
apn
3sks
−−
−−
−−
−−
−−
The infinitive with zu can also be realized as an infix of the verb. In this case, the verb
is tagged as VVIZU. Moreover, it is projected to VXINF with the grammatical function
HD:
69
SIMPX
509
−
−
−
MF
508
MOD
MOD
OA
PX
507
−
HD
NX
505
VC
506
−
C
500
HD
HD
ADJX
501
−
ADVX
502
HD
um
neben
0
NX
503
HD
neuen
Baugesetzen
1
−
auch
2
3
VXINF
504
HD
mehr
4
HD
Mitspracherechte
einzufordern
5
6
.
7
8
KOUI
APPR
ADJA
NN
ADV
PIAT
NN
VVIZU
$.
−−
d
dpn
dpn
−−
***
apn
−−
−−
Beside the examples above, the infinitive with zu occurs in optional (in most cases
with um zu) and obligatory infinitive clauses.
SIMPX
515
−
−
−
−
NF
514
OS
MF
512
ON
OS−MOD
SIMPX
513
OD
OPP
−
PX
508
−
HD
C
500
NX
501
NX
502
NX
503
NX
504
−
HD
HD
HD
HD
Wenn
Angehörige
0
es
1
sich
2
zur
3
MF
510
HD
OA
VXFIN
505
Lebensaufgabe
HD
VXINF
507
−
machen
5
VC
511
NX
506
HD
4
−
VC
509
,
6
HD
den
7
HD
Kranken
8
−
zu
9
kontrollieren
10
11
KOUS
NN
PPER
PRF
APPRART
NN
VVFIN
$,
ART
NN
PTKZU
VVINF
−−
np*
asn3
dp*3
dsf
dsf
3pis
−−
asm
asm
−−
−−
SIMPX
513
−
−
−
−
NF
512
OS
SIMPX
511
−
MF
507
VC
508
OD
HD
C
500
NX
501
−
HD
um
Freunden
0
C
503
−
zu
1
−
MF
509
VXINF
502
HD
−
−
sagen
2
,
3
−
daß
5
ON
OA
NX
504
NX
505
HD
ihr
4
VC
510
VXFIN
506
HD
Zug
6
HD
HD
Verspätung
7
hat
8
9
KOUI
NN
PTKZU
VVINF
$,
KOUS
PPOSAT
NN
NN
VAFIN
−−
dpm
−−
−−
−−
−−
nsm
nsm
asf
3sis
70
Infinitive clauses can consist of only one verb complex:
SIMPX
518
−
−
FKOORD
517
KONJ
−
KONJ
FKONJ
516
−
−
−
FKONJ
514
NF
515
−
−
OS
MF
512
SIMPX
513
OA
VF
507
LK
508
ON
HD
NX
500
HD
−
PX
509
−
VXFIN
501
HD
NX
502
HD
Er
V−MOD
−
wendet
0
den
1
NX
503
HD
−
Blick
2
von
3
VC
511
HD
VXFIN
504
HD
der
4
LK
510
HD
Wand
5
VC
505
HD
und
6
VPT
fängt
7
an
8
VXINF
506
HD
−
zu
9
erzählen
10
.
11
12
PPER
VVFIN
ART
NN
APPR
ART
NN
KON
VVFIN
PTKVZ
PTKZU
VVINF
$.
nsm3
3sis
asm
asm
d
dsf
dsf
−−
3sis
−−
−−
−−
−−
4.7.5
Coherency and Incoherency of Verbal Constructions
The notion of coherency attributed to Bech (1955 57) covers the relation of dependency
between adjacent verbal elements, i.e. the relation of subcategorization between a verb
and a non-finite verbal complement. Kiss (1995) calls this relation infinitive Komplementation (non-finite complementation). Bech (1955 57) distinguishes between three different
modi of obligatory and optional coherency:
1. verbs constructing coherently and incoherently, e.g. versprechen, versuchen
coherent, extraposition possible:
a. [wie er mit kritischen politischen Gegenpositionen umzugehen versteht]
incoherent, extraposition:
b. [wie er versteht,][mit kritischen politischen Gegenpositionen umzugehen]
2. verbs constructing only coherently, e.g. wollen, möchten
coherent, no extrapostion possible:
a. [wie er mit kritischen politischen Gegenpositionen umgehen will]
b.*[wie er will mit kritischen politischen Gegenpositionen umgehen]
3. verbs constructing only incoherently, e.g. überreden, überzeugen
incoherent, extraposition obligatory:
a. [wie er sie überredet,][mit kritischen politischen Gegenpositionen umzugehen]
b.*[wie er sie [mit kritischen politischen Gegenpositionen umzugehen] überredet]
71
Coherent and incoherent constructions of verbs are annotated differently. In case of
coherency, the verbal complement is part of the verb complex. In the clause wie er mit
kritischen politischen Gegenpositionen umzugehen versteht, for instance, the infinitive
with zu is the verbal object of the finite verb. While in case of incoherency, the verbal complement is annotated as a sentential complement, i.e., mit kritischen politischen
Gegenpositionen umzugehen in the clause wie er sie überredet, mit kritischen politischen
Gegenpositionen umzugehen is a sentential object in NF.
We define that a construction is incoherent, if extraposition in NF is possible. That
is, whenever it is possible to shift the infinitival complement together with a constituent
of MF, which it subcategorizes for, into NF, these elements are annotated as sentential objects. Therefore, the coherent example above (wie er mit kritischen politischen
Gegenpositionen umzugehen versteht) is annotated with a sentential object in MF since
extraposition is possible (cf. the incoherent example 1.b.).
SIMPX
513
−
−
−
MF
512
ON
OS
SIMPX
511
−
−
MF
510
OPP
PX
509
−
HD
NX
506
−
C
500
NX
501
−
HD
wie
ADJX
502
mit
1
HD
kritischen
2
HD
ADJX
503
HD
er
0
−
politischen
3
Gegenpositionen
4
VC
507
VC
508
HD
HD
VXINF
504
VXFIN
505
HD
HD
umzugehen
5
versteht
6
7
KOUS
PPER
APPR
ADJA
ADJA
NN
VVIZU
VVFIN
−−
nsm3
d
dpf
dpf
dpf
−−
3sis
If a complement of the verb within the sentential object is located out of the sentence boundaries, e.g. in the C-field, the secondary edge label refcontr gives additional
information about the dependency relation (cf. 3.4.6).
4.7.6
AcI Constructions
AcI (accusativus cum infinitivo) verbs are a small group of verba sentiendi (e.g. sehen,
hören, fühlen, spüren) which subcategorize for an accusative and an infinitive. The verbs
lassen, machen, heißen have a modal verb like reading in which they also select an
accusative and an infinitive.
The infinitive itself subcategorizes for complements with respect to its valency but its
subject is realized by an accusative which is the direct object of the AcI verb.
Since AcI constructions are coherent infinitive constructions in which extraposition is
not possible (cf. (Eisenberg 1999 2001), p.355), the AcI is not annotated as a sentential
object (* wenn man nur noch hört das eigene Herz schlagen). The infinitive as the verbal
object of the AcI verb is located in the verb complex and the accusative is realized as OA
in MF.
72
SIMPX
510
−
−
−
MF
509
ON
MOD
MOD
OA
NX
507
−
C
500
NX
501
−
HD
Wenn
ADVX
502
ADVX
503
HD
HD
man
0
nur
1
VC
508
−
HD
ADJX
504
HD
noch
2
das
3
eigene
4
Herz
5
OV
HD
VXINF
505
VXFIN
506
HD
HD
schlagen
6
hört
7
8
KOUS
PIS
ADV
ADV
ART
ADJA
NN
VVINF
VVFIN
−−
ns*
−−
−−
asn
asn
asn
−−
3sis
As a consequence of this analysis, we annotate two accusative objects (OA) if the
AcI construction comprises a transitive infinitive verb such as beenden in the following
example. Uns functions as its subject and die Diskussion as its direct object. Both are
in accusative case and both are labelled OA.
SIMPX
509
−
−
LK
506
HD
VXFIN
500
HD
Lassen
−
MF
507
OA
ON
OA
NX
501
HD
NX
502
HD
NX
503
−
VC
508
OV
MOD
HD
ADVX
504
HD
VXINF
505
HD
jetzt
beenden
Sie
uns
die
Diskussion
VVFIN
PPER
PPER
ART
NN
ADV
VVINF
3pis
np*3
ap*1
asf
asf
−−
−−
0
4.7.7
1
2
3
4
5
6
Imperatives
Imperative verbs have only one singular and one plural form and are not inflected concerning the grammatical category person. Their form corresponds to second person singular
and plural verbs which are tagged as VVIMP or VAIMP.
Warte mal!
instead of
Wartest du mal?
(warte/VVIMP:s)
(wartest/VVFIN:2sis)
It is important to keep apart imperative sentences from imperative verbs. An imperative sentence does not need to comprise an imperative verb form as is shown in the
following examples
Warten Sie mal bitte!
Bitte warten!
(warten/VVFIN:3pis)
(warten/VVINF:–)
73
SIMPX
506
−
−
MF
505
OA
LK
503
HD
NX
504
HD
VXFIN
500
HD
−
NX
501
HD
(
vgl.
$(
VVIMP
NN
CARD
$(
−−
s
asf
−−
−−
0
Seite
NX
502
HD
1
32
2
)
3
4
Normally imperative verbs are lacking the subject, but the addressed person can also
be mentioned to stress the utterance:
SIMPX
504
−
DM
502
−
LK
503
HD
NX
500
HD
VXFIN
501
HD
Maikäfer
flieg
0
...
1
2
NN
VVIMP
$(
nsm
s
−−
4.7.8
Particle Verbs
Separable verb particles are tagged as PTKVZ and annotated with the edge label VPT:
SIMPX
511
−
−
MF
510
ON
OD
LK
507
HD
ADVX
500
−
NX
501
HD
−
Auch
−
Vertreter
1
−
−
VXFIN
503
HD
der
2
NX
508
HD
NX
502
HD
die
0
−
VF
509
NX
506
−
−
ADJX
504
HD
AfB
3
HD
VC
505
HD
stimmten
4
den
5
VPT
86
6
Millionen
7
zu
8
.
9
10
ADV
ART
NN
ART
NE
VVFIN
ART
CARD
NN
PTKVZ
$.
−−
npm
npm
gsf
gsf
3pit
dpf
−−
dpf
−−
−−
74
In verb-final clauses, the particle verb occurs unseparated within the verb complex:
SIMPX
510
−
−
−
−
VF
506
LK
507
ON
HD
MOD
OD
MOD
OV
VXFIN
501
ADVX
502
NX
503
ADVX
504
VXINF
505
HD
HD
HD
HD
NX
500
HD
Rußland
MF
508
wollte
0
VC
509
−
bislang
1
HD
einer
2
UN−Resolution
3
nur
4
zustimmen
5
6
NE
VMFIN
ADV
ART
NN
ADV
VVINF
nsn
3sit
−−
dsf
dsf
−−
−−
4.7.9
Verbs with Predicate
Typically, the complement type PRED (predicate) occurs with verbs like sein, haben,
scheinen, aussehen, sich anhören, klingen, etc. PRED is annotated, if the following
conditions apply:
• if it is not possible to determine the case of the constituent in question properly
(e.g. gut in Das ist gut.)
• if the constituent in question actually predicates the subject, i.e. the subject is
characterized as having the property expressed by PRED (e.g. in Die Ursache war
unklar. Die Ursache is characterized by the property of being unclear)
• many PRED verbs are raising-verbs (subject without theta-role)
• if als-phrases are selected by the verb they are labelled as PRED (e.g. Unter dem
Motto Kino-Extrem agiert der Regisseur als Filmjockey.)
SIMPX
510
−
−
−
VF
509
V−MOD
PX
508
−
HD
NX
LK
505
Unter
APP
HD
ON
PRED
NX
NX
VXFIN
NX
NX
501
HD
dem
0
507
APP
500
−
MF
506
HD
Motto
1
502
"
2
HD
Kino−Extrem
3
503
"
4
−
agiert
5
der
6
504
HD
−
Regisseur
7
HD
als
8
Filmjockey
9
10
APPR
ART
NN
$(
NN
$(
VVFIN
ART
NN
KOKOM
NN
d
dsn
dsn
−−
dsn
−−
3sis
nsm
nsm
−−
nsm
75
Some examples for verbs that take predicates: recht sein, recht haben, leid tun, frei
sein, fertig sein, sich gut/schlecht treffen, gut/schlecht finden, etc.
PRED verbs have to be distinguished carefully from verbs occurring with ordinary
modifiers (V-MOD) such as gut passen.
With respect to topological fields, note that PRED usually marks the border between
MF and NF, i.e., whatever constituent occurs on the right hand side of PRED belongs to
NF. In general, this constituent is an adjunct which PRED does not subcategorize for:
SIMPX
508
−
−
−
VF
504
LK
505
ON
HD
NX
500
HD
VXFIN
501
NF
507
V−MOD
NX
502
ADVX
503
HD
ist
0
MF
506
PRED
HD
Das
−
HD
Politik
1
hier
2
3
PDS
VAFIN
NN
ADV
nsn
3sis
nsf
−−
SIMPX
509
−
−
−
−
NF
508
V−MOD
VF
504
LK
505
ON
HD
NX
500
HD
VXFIN
501
HD
es
PX
507
−
HD
ADJX
502
NX
503
HD
ist
0
MF
506
PRED
−
kalt
1
an
2
HD
diesem
3
Tag
4
5
PPER
VAFIN
ADJD
APPR
PDAT
NN
nsn3
3sis
−−
d
dsm
dsm
But there are exceptions in which PRED does not necessarily constitute the border
between MF and NF:
• Another constituent may occur between PRED and VC, for instance, if an ambiguous modifier follows PRED.
76
R−SIMPX
511
−
−
−
MF
510
V−MOD
C
505
PRED
PX
506
ON
NX
507
−
HD
NX
500
−
an
0
HD
diesem
1
HD
−
Abend
der
3
HD
NX
503
HD
2
VC
509
HD
ADJX
502
−
das
PX
508
−
NX
501
HD
MOD
HD
schönste
Platz
4
5
im
6
HD
All
7
VXFIN
504
war
8
.
9
10
PRELS
APPR
PDAT
NN
ART
ADJA
NN
APPRART
NN
VAFIN
$.
nsn
d
dsm
dsm
nsm
nsm
nsm
dsn
dsn
3sit
−−
• PRED subcategorizes for the constituent that follows it. Complements of PREDs
are always attached to a field since they are assigned a grammatical function within
the sentence structure (cf. 8.1):
SIMPX
510
−
−
−
VF
508
MF
509
ON
PRED
NX
505
LK
506
HD
−
NX
500
HD
meiner
0
−
VXFIN
502
−
Einer
PX
507
HD
NX
501
HD
FOPP
ADJX
503
HD
Freunde
1
NX
504
HD
wurde
2
HD
HD
süchtig
3
nach
4
Nachrichten
5
.
6
7
PIS
PPOSAT
NN
VAFIN
ADJD
APPR
NN
$.
nsm
gpm
gpm
3sit
−−
d
dpf
−−
SIMPX
507
−
−
−
VF
504
LK
505
ON
HD
NX
500
HD
PRED
VXFIN
501
HD
ich
FOPP
ADJX
502
HD
bin
0
MF
506
froh
1
PX
503
HD
darum
2
3
PPER
VAFIN
ADJD
PROP
ns*1
1sis
−−
−−
• Because of the word order rule that pronouns in MF have to precede other constituents, PRED might not be the last element in MF if it is a pronoun:
77
SIMPX
510
−
−
VF
−
LK
506
−
MF
507
VC
508
509
ON
HD
PRED
MOD
OV
HD
NX
VXFIN
NX
ADVX
VXINF
VXINF
500
501
HD
502
HD
Bravo
503
HD
kann
0
HD
es
1
504
505
HD
nicht
HD
gewesen
2
3
sein
4
.
5
6
NN
VMFIN
PPER
PTKNEG
VAPP
VAINF
$.
nsf
3sis
nsn3
−−
−−
−−
−−
SIMPX
507
−
−
−
VF
504
LK
505
ON
HD
NX
500
MF
506
PRED
VXFIN
501
HD
NX
502
HD
er
ADVX
503
HD
war
0
MOD
HD
es
1
nicht
2
3
PPER
VAFIN
PPER
PTKNEG
nsm3
3sit
nsn3
−−
4.7.10
Modal Verbs
Modal verbs are always tagged as VMFIN or VMINF regardless of their use as an auxiliary
or a main verb. If a modal verb functions as an auxiliary verb, it is projected like any
other auxiliary verb. If a modal verb is the main verb of a sentences, verbal modifiers
refer to the modal verb in the same way as they refer to other main verbs:
SIMPX
508
−
−
−
VF
504
LK
505
MF
506
VC
507
ON
HD
OA
OV
NX
500
HD
"
−
NX
502
HD
Die
0
VXFIN
501
−
wollten
1
HD
die
2
VXINF
503
HD
BLG
3
schonen
4
.
5
"
6
7
$(
PDS
VMFIN
ART
NE
VVINF
$.
$(
−−
np*
3pit
asf
asf
−−
−−
−−
78
SIMPX
512
−
−
LK
509
−
MF
510
HD
VXFIN
500
HD
Hätte
VC
511
ON
OD
OA
MOD
NX
501
NX
502
NX
503
ADVX
504
HD
HD
HD
sie
0
sich
1
HD
das
2
OA−MOD
V−MOD
OV
HD
NX
505
ADVX
506
VXINF
507
VXINF
508
HD
HD
HD
HD
nicht
3
alles
4
vorher
5
überlegen
6
können
7
?
8
9
VAFIN
PPER
PRF
PDS
PTKNEG
PIS
ADV
VVINF
VMINF
$.
3skt
nsf3
ds*3
asn
−−
asn
−−
−−
−−
−−
SIMPX
508
−
−
−
MF
507
ON
OPP
C
504
V−MOD
PX
505
−
PX
500
HD
Warum
HD
NX
501
HD
0
Daewoo
VC
506
HD
NX
502
HD
1
nach
2
Bremen
VXFIN
503
HD
3
mußte
PWAV
NE
APPR
NE
VMFIN
−−
ns*
d
dsn
3sit
4
79
Chapter 5
Attachment Principles for Phrases
5.1
Attachment to Fields
Phrases are attached to the topological field in which they occur. Their edge labels denote
their grammatical function within the sentence structure. In LK and VC there can only
occur verb forms, separable verbal prefixes, or infinitive particles. LK and VC mark the
beginning and the end of MF (cf. 3.2).
5.2
Attachment of Ambiguous Complements
The partially free word order and the morphological properties of German can cause
ambiguity concerning the grammatical function of a constituent. In the following example,
the syntactic structure does not give any information about case assignment. Both noun
phrases can be identified as ON or OA:
SIMPX
509
−
−
−
VF
508
OA
NX
507
HD
−
PX
504
−
HD
NX
500
−
NX
501
HD
Ein
−
Bad
0
in
1
MF
506
HD
ON
VXFIN
502
HD
der
2
LK
505
HD
Menge
3
NX
503
−
verhindert
4
HD
das
5
Sicherheitsgitter
6
.
7
8
ART
NN
APPR
ART
NN
VVFIN
ART
NN
$.
asn
asn
d
dsf
dsf
3sis
nsn
nsn
−−
Headlines like the following are lacking the finite verb. Therefore, in the first example
it cannot be decided if it is an active or a passive construction, i.e., if the noun phrase is
ON or OA. The second example is an active construction, but again the noun phrase can
be both, ON or OA:
80
SIMPX
504
−
−
MF
502
VC
503
ON
HD
NX
500
VXINF
501
HD
HD
Kriegsverbrecher
verurteilt
0
1
NN
VVPP
npm
−−
SIMPX
504
−
−
MF
502
VC
503
OA
HD
NX
500
VXINF
501
HD
HD
Prachtwicken
gucken
0
.
1
2
NN
VVINF
$.
apf
−−
−−
Since we do not assign specific edge labels for ambiguous complements, we formulate
the following preference principle for case assignment:
Preference principle for case assignment:
If case assignment is ambiguous, we decide on the more plausible grammatical function
and on the more plausible sequence of grammatical functions respectively. The main
criteria for the decision are the unmarked word order and the semantic content.
Therefore, in the first example above, OA appears in VF whereas ON has its position
in MF. For elliptical headlines, we assume a passive construction if the verb in VC is a
past participle and an active construction if the verb in VC is an infinitive (cf. 4.7.2 and
7.4).
5.3
Modifier Attachment
Modifiers either modify one specific constituent or more than one constituent. The scope
of modification can even range over the whole sentence structure. Therefore, they are
either unambiguous or ambiguous. An unambiguous constituent that modifies just one
other constituent within a tree structure is either adjacent or discontinuous. In the
first case, it is immediately attached to the constituent which it modifies, concerning
the attachment rules for phrases. In the second case, the dependency, which can even go
beyond the border of topological fields, is indicated by X-MOD edge labels, which express
the non-ambiguity of the modifier (e.g. OA-MOD is the modifier of OA). Thus, edge
labels like OA-MOD, V-MOD, OPP-MOD, MOD-MOD, etc. express that the respective
constituent modifies only one other constituent in the sentence (OA, V, OPP, a modifier,
etc.) which is not adjacent:
81
SIMPX
511
−
−
−
−
VF
510
OA−MOD
PX
506
LK
507
−
HD
HD
NX
500
−
Für
VXFIN
501
HD
diese
0
MF
508
HD
Behauptung
1
ON
MOD
NX
502
ADVX
503
HD
hat
2
OA
OV
NX
504
HD
Beckmeyer
3
VC
509
−
bisher
keinen
4
5
VXINF
505
HD
HD
Nachweis
6
geliefert
.
7
8
9
APPR
PDAT
NN
VAFIN
NE
ADV
PIAT
NN
VVPP
$.
a
asf
asf
3sis
nsm
−−
asm
asm
−−
−−
If a modifying constituent is ambiguous (i.e. it modifies more than one constituent,
the entire sentence, or a constituent that occurred in previous sentences), it is attached
to its topological field and given the ambiguous edge label MOD to preserve ambiguity.
In the following example an der Uni either modifies the accusative object den Entwicklungsprozeß or the verb fortsetzen:
SIMPX
513
−
−
−
VF
511
V−MOD
ON
ADJX
507
KONJ
KONJ
ADJX
500
MOD
HD
energisch
1
NX
503
HD
will
2
VC
510
−
VXFIN
502
HD
und
PX
509
HD
ADJX
501
HD
0
OA
LK
508
−
Phantasievoll
−
MF
512
NX
504
−
er
3
HD
NX
505
HD
den
4
OV
−
Entwicklungsprozeß
5
an
6
der
7
VXINF
506
HD
HD
Uni
8
fortsetzen
9
10
ADJD
KON
ADJD
VMFIN
PPER
ART
NN
APPR
ART
NN
VVINF
−−
−−
−−
3sis
nsm3
asm
asm
d
dsf
dsf
−−
We formulate the following definitions for MOD and X-MOD:
Definition of MOD:
A constituent is called MOD, if it cannot be assigned a more specific label, either
because it is ambiguous or because there is no more specific label (e.g. for sentence
modifiers or for constituents that refer to some sentence external expression). Sometimes
it is difficult to determine whether a modifier is definite or not. In cases of doubt, modifiers
are marked as ambiguous (MOD) rather than as definite modifiers.
Definition of X-MOD:
X is a variable that can be replaced by labels for syntactic categories like OA, OPP,
MOD, V. X-MOD marks long-distance modification which is unambiguous, e.g. relative
clauses (Aber es gäbe (intelligente Lösungen OA), (die kein Geld kosten OA-MOD)).
Typical MODs and V-MODs:
Generally, modifying subclauses (e.g. Katastrophenstimmung herrscht erst, [wenn
82
nichts mehr zu verheimlichen ist] (MOD).) are MOD because they modify the complete main clause. Modifying particles and adverbs like da, dann, auch, eigentlich, ja,
vielleicht, auch, natürlich usually show attachment ambiguity and therefore are annotated as MOD. Only if they unambiguously express the modification of the verb (e.g. Das
Buch liegt da. or Er geht auch.) they carry the edge label V-MOD. Pronominal adverbs
(PROP) like dabei, dafür, trotzdem, deswegen, hierauf, etc. are either ambiguous (e.g.
Dabei (MOD) erscheinen Sie in anderen Verlagen.) or unambiguous [e.g. Er achtet dabei
(V-MOD) auf alles.). Non-pronominal adverbs such as vorher, später, etc. in most cases
give temporal or local information. Thus, they are rather V-MOD than MOD.
5.3.1
Modifier Attachment in the Initial Field
Since only one constituent is allowed in the initial field, all elements preceding and following the head are attached as premodifiers (low attachment) or postmodifiers (high
attachment) according to the attachment rules explained in 4.1.
SIMPX
513
−
−
−
VF
511
MOD
MF
512
ON
PX
509
−
PX
510
refint
−
HD
−
NX
506
HD
ADVX
500
HD
Auch
PRED
−
NX
501
HD
HD
509
LK
507
HD
ADVX
502
HD
VXFIN
503
HD
NX
504
−
HD
ADJX
505
HD
HD
für
Rumänien
ist
der
Papst−Besuch
ADV
APPR
NE
ADV
VAFIN
ART
NN
APPR
ADJA
NN
$.
−−
a
asn
−−
3sis
nsm
nsm
d
dsf
dsf
−−
0
1
5.3.2
2
selbst
NX
508
−
3
4
5
von
6
großer
7
8
Bedeutung
9
.
10
Attachment across Punctuation Marks
The punctuation marks : and - and ... separate a syntactic construction within a unit
unless there is no syntactic dependency relation between the two parts (cf. 3.4.5) like in
the following:
SIMPX
NX
508
−
−
VF
509
−
HD
−
LK
505
ON
HD
NX
VXFIN
500
HD
NX
506
VC
501
HD
ASB
502
503
:
2
504
HD
Tag
3
HD
ADJX
HD
ein
1
−
NX
VPT
lädt
0
507
−
der
4
offenen
5
Tür
6
7
NN
VVFIN
PTKVZ
$.
NN
ART
ADJA
NN
nsm
3sis
−−
−−
nsm
gsf
gsf
gsf
83
NX
502
−
NX
500
−
HD
Sein
HD
ADJX
501
HD
Zuhause
0
:
1
stilvolles
2
Entertainment
.
3
4
5
PPOSAT
NN
$.
ADJA
NN
$.
nsn
nsn
−−
nsn
nsn
−−
Attachment is necessary if the part following the punctuation mark has a grammatical
function within the sentence structure:
SIMPX
512
−
−
−
NF
511
OS
SIMPX
510
−
−
−
VF
505
LK
506
VF
507
LK
508
ON
HD
ON
HD
NX
500
VXFIN
501
HD
NX
502
HD
Er
meinte
:
1
−
PRED
VXFIN
503
HD
0
MF
509
NX
504
HD
das
−
ist
4
HD
meine
5
Geschichte
6
’
7
.
"
2
3
8
9
PPER
VVFIN
$.
$(
PDS
VAFIN
PPOSAT
NN
$(
$.
$(
10
nsm3
3sit
−−
−−
nsn
3sis
nsf
nsf
−−
−−
−−
SIMPX
516
−
−
−
−
NF
515
V−MOD
PX
514
−
KONJ
−
−
KONJ
PX
512
−
KOORD
500
VF
508
LK
509
ON
HD
NX
501
−
Doch
Zweifel
−
NX
503
HD
blieben
1
HD
NX
511
HD
HD
0
−
NX
510
VXFIN
502
HD
PX
513
HD
−
2
sowohl
3
bei
4
Joergensen
5
HD
ADVX
504
ADVX
505
HD
HD
selbst
6
als
7
−
auch
8
−
NX
506
in
9
HD
der
10
NX
507
−
Redaktion
11
HD
der
12
taz
13
.
14
15
KON
NN
VVFIN
$(
KON
APPR
NE
ADV
KON
ADV
APPR
ART
NN
ART
NE
$.
−−
npm
3pit
−−
−−
d
dsm
−−
−−
−−
d
dsf
dsf
gsf
gsf
−−
5.3.3
Ambiguous Modifiers in Isolated Phrases
Since isolated phrases (cf. 3.4.5) do not consist of topological fields, ambiguous modifiers
(MOD) have to be attached to the phrase itself. The isolated phrase is projected one
level higher and the modifier is attached on this higher level. Thus, the information about
ambiguity can be preserved even without topological fields or explicit MOD labelling, just
by the existence of yet another projection level of the phrase.
The overall attachment strategy has been chosen in order to keep syntactic structure
84
flat and to be able to preserve attachment ambiguity where necessary.
In the following examples, so may refer to something that is implicit or has been
mentioned before:
NX
502
−
HD
ADVX
500
NX
501
HD
HD
so
Winkler
0
1
ADV
NE
−−
nsf
If there is more than one ambiguous modifier in an isolated phrase, all of them are
attached on the next higher level. The mother node of this isolated phrase is marked
with the node label of the modified phrase.
NX
503
−
−
ADVX
500
ADVX
501
HD
HD
Zunächst
HD
NX
502
HD
natürlich
0
Durcheinander
1
.
2
3
ADV
ADV
NN
$.
−−
−−
nsn
−−
NX
504
−
−
ADVX
500
ADVX
501
HD
HD
vielleicht
HD
−
mal
0
ADVX
503
HD
ein
1
−
NX
502
HD
Mini−Hit
2
da
3
4
ADV
ADV
ART
NN
ADV
−−
−−
nsm
nsm
−−
85
Chapter 6
The Annotation of Sentences
The approach of topological fields supports the flat clustering principle inasmuch MF
and NF allow for more than one constituent being attached to the same field node. The
field nodes form a level of annotation between the phrase level and the sentence level.
The last step to complete a sentence structure is to attach the field nodes to the highest
annotation level of the whole structure: the root node.
In the following sections, the annotation of sentence structures will be demonstrated.
6.1
Sentence Initial Fields
6.1.1
The C-Field in Verb-Final Clauses
The C-field (complementizer field) is the field for subordinating conjunctions KOUS (e.g.
daß, wenn, da, weil, ob), KOUI (e.g. um (+zu)), relative pronouns (PRELS), interrogative
(PWAV) pronouns and (complex) interrogative or relative phrases. Thus, it only occurs
in verb-final clauses, except for comparison clauses with the conjunction als.
In case of a conjunction, we directly project to the C-field:
SIMPX
508
−
−
−
MF
506
MOD
VC
507
ON
OV
refvc
C
500
ADVX
501
−
HD
Wenn
da
0
NX
502
HD
was
1
OV
503
HD
VXINF
503
VXINF
504
VXFIN
505
HD
HD
HD
gebucht
2
worden
3
ist
4
5
KOUS
ADV
PIS
VVPP
VAPP
VAFIN
−−
−−
***
−−
−−
3sis
There are conjunctions in German which consist of two elements (e.g. so daß and als
ob). Both of them are also directly attached to the C-field, while none of them carries a
head label.
86
SIMPX
510
−
−
−
MF
509
ON
V−MOD
PRED−MOD
PRED
PX
508
−
HD
NX
506
VC
507
−
C
500
−
NX
501
−
so
ADVX
502
−
daß
HD
der
0
HD
Maschinenpark
HD
HD
ADJX
503
ADJX
504
HD
heute
3
für
4
VXFIN
505
HD
lukrative
5
Sonderanfertigungen
6
HD
unbrauchbar
7
ist
8
.
1
2
KOUS
KOUS
ART
NN
ADV
APPR
ADJA
NN
ADJD
VAFIN
9
$.
10
−−
−−
nsm
nsm
−−
a
apf
apf
−−
3sis
−−
Since C generally does not contain more than one constituent, the adverb auch in the
following example is not supposed to occur in the C-field together with the conjunction
wenn. The wenn-clause is annotated as the modifier of the adverbial phrase auch, i.e.,
the adverbial phrase subcategorizes for the verb-final clause.
ADVX
516
HD
−
SIMPX
515
−
−
−
MF
514
ON
MOD
OA
NX
512
−
HD
−
NX
509
KONJ
ADVX
500
HD
auch
,
0
−
−
NX
502
NX
503
NX
504
−
HD
HD
HD
wenn
man
2
wie
3
ARD
4
und
5
HD
DP
510
KONJ
C
501
1
V−MOD
NX
513
ADJX
505
HD
ZDF
6
VC
511
HD
relativ
7
viele
8
Teams
9
OV
HD
ADVX
506
VXINF
507
VXFIN
508
HD
HD
HD
überall
10
postieren
11
kann
12
13
ADV
$,
KOUS
PIS
KOKOM
NE
KON
NE
ADJD
PIDAT
NN
ADV
VVINF
VMFIN
−−
−−
−−
ns*
−−
nsf
−−
nsn
−−
apn
apn
−−
−−
3sis
If the constituent in the C-field is a pronoun or a complex phrase, it is first projected
to the phrase level and then projected to the C-field. The edge label below the C-Field
denotes the grammatical function of this constituent.
SIMPX
508
−
−
C
505
−
MF
506
ON
MOD
NX
500
ADVX
501
HD
HD
Wieviel
V−MOD
PRED
ADJX
502
HD
monatlich
1
HD
ADJX
503
HD
da
0
VC
507
HD
fällig
2
VXFIN
504
wird
3
4
PWS
ADV
ADJD
ADJD
VAFIN
nsn
−−
−−
−−
3sis
87
R−SIMPX
510
−
−
C
508
MF
509
FOPP
ON
PX
505
MOD
MOD
NX
506
−
HD
−
deren
HD
HD
ADJX
501
HD
0
VC
507
−
NX
500
−
zu
−
HD
Ablaß
1
die
2
tonale
3
Ebene
4
ADVX
502
ADVX
503
VXFIN
504
HD
HD
HD
natürlich
5
nicht
6
ausreicht
7
8
APPR
PRELAT
NN
ART
ADJA
NN
ADV
PTKNEG
VVFIN
d
gp*
dsm
nsf
nsf
nsf
−−
−−
3sis
6.1.2
The C-Field in Verb-Second Clauses
Only comparison clauses with als allow for a C-field and a left sentence bracket in the
same clause:
SIMPX
506
−
−
−
LK
504
HD
C
500
−
ON
VXFIN
501
HD
als
PRED
NX
502
HD
sei
0
MF
505
das
1
NX
503
−
HD
deren
2
Pflicht
3
4
KOUS
VAFIN
PDS
PDAT
NN
−−
3sks
nsn
gsf
nsf
6.1.3
The KOORD-Field in all Clause Types
The KOORD-field is optionally the left-most field of all clause types (V-1, V-2, V-end).
Therefore, it can only occur at the beginning of a syntactic unit (cf. 3.4.3).
For verb-second clauses, it can be regarded as an alternative field to the PARORDfield. The KOORD-field contains coordinative particles like und, oder, aber, etc. (cf.
Höhle (1986)). Here are two examples of different clause types:
88
−
−
SIMPX
513
−
−
−
MF
512
V−MOD
MOD
OPP
PX
511
−
KOORD
500
−
Und
VF
507
ON
LK
508
HD
NX
501
HD
VXFIN
502
HD
ADVX
503
HD
war
früher
Koring
HD
NX
509
−
ADVX
504
HD
VC
510
OV
HD
ADJX
505
HD
in
schiefes
KON
NE
VAFIN
ADV
ADV
APPR
ADJA
NN
VVPP
−−
nsm
3sit
−−
−−
a
asn
asn
−−
0
1
2
einmal
VXINF
506
HD
3
4
5
6
Licht
7
geraten
8
SIMPX
507
−
−
−
LK
505
MF
506
HD
KOORD
500
VXFIN
501
−
HD
Oder
MOD
PRED
NX
502
ADVX
503
ADJX
504
HD
ist
0
ON
HD
Bremerhaven
1
HD
nicht
2
günstiger
3
?
4
5
KON
VAFIN
NE
PTKNEG
ADJD
$.
−−
3sis
nsn
−−
−−
−−
6.1.4
The PARORD-Field in Verb-Second Clauses
PARORD is an alternative field to KOORD for verb-second clauses only.
PARORD expressions are denn, weil1 :
Typical
SIMPX
509
−
−
−
−
−
VF
508
ON
NX
505
−
LK
506
HD
HD
PARORD
500
ADVX
501
VXFIN
502
−
HD
HD
Denn
auch
0
die
1
MF
507
OPP
gehen
2
PX
503
VC
504
HD
VPT
davon
3
aus
4
5
KON
ADV
PDS
VVFIN
PROP
PTKVZ
−−
−−
np*
3pis
−−
−−
1
weil can occur in verb-second and in verb-final clauses. In the first case, it is in the PARORD-field,
in the latter case, it belongs to the C-field.
89
6.1.5
Resumptive Constructions: The LV-Field
Resumptive constructions are analysed as suggested by Höhle (1986) and Kathol (1995),
by using the field LV (Linksversetzung) which is located to the left of VF. In general,
the LV-field is not restricted to one constituent. The typical feature of a resumptive
construction is that there is a (pronominal) constituent somewhere in the sentence, on
the right hand side of the LV-field, which refers back to the expression within the LV-field.
Therefore, we use the X-MOD label to indicate this kind of long-distance dependency.
SIMPX
520
−
−
−
−
−
LV
519
ON−MOD
PX
518
KONJ
KONJ
PX
516
PX
517
−
HD
−
HD
NX
510
NX
511
−
HD
−
ADJX
500
ADJX
501
HD
Vom
Einzelgänger
zum
1
2
LK
513
HD
MOD
MOD
MOD
MOD
OV
HD
VXFIN
503
ADVX
504
ADVX
505
ADVX
506
ADVX
507
VXINF
508
VXINF
509
HD
HD
HD
HD
HD
HD
HD
ja
auch
HD
stilbildenden
3
VF
512
ON
NX
502
HD
introvertierten
0
HD
Popstar
4
,
5
das
6
muß
7
MF
514
8
9
VC
515
erst
10
einmal
11
verkraftet
12
werden
13
14
APPRART
ADJA
NN
APPRART
ADJA
NN
$,
PDS
VMFIN
ADV
ADV
ADV
ADV
VVPP
VAINF
dsm
dsm
dsm
dsm
dsm
dsm
−−
nsn
3sis
−−
−−
−−
−−
−−
−−
SIMPX
515
−
−
−
−
−
LV
514
FOPP−MOD
SIMPX
513
−
−
−
MF
508
ON
KOORD
500
−
Doch
C
501
NX
502
−
HD
wie
0
es
1
VC
509
VF
510
OV
HD
VXINF
503
VXFIN
504
HD
HD
weitergehen
2
FOPP
,
4
HD
ON
NX
507
HD
darüber
5
MF
512
VXFIN
506
HD
soll
3
PX
505
LK
511
−
herrscht
6
HD
kein
7
Konsens
8
.
9
10
KON
KOUS
PPER
VVINF
VMFIN
$,
PROP
VVFIN
PIAT
NN
$.
−−
−−
nsn3
−−
3sis
−−
−−
3sis
nsm
nsm
−−
Grammatical functions within a LV-construction are assigned according to the following principle:
• The LV-constituent is licensed by some (pronominal) constituent within the core
sentence. The core sentence exceeds from VF to NF. Therefore, the licensing constituent is considered to be modified by the constituent within the LV-field.
For instance, ON-MOD is licensed by ON like in the first example above, which is
also in strong accordance with the assumption that the original position of the subject in
verb-second clauses is VF.
In constructions with wenn ... dann ..., the wenn-clause, which is semantically a precondition to the dann-clause, is in the LV-field in correlation with dann. Therefore, dann
(MOD) refers back to the wenn-clause (MOD-MOD):
90
15
SIMPX
519
−
−
−
−
LV
518
MOD−MOD
SIMPX
516
−
MF
517
−
−
MF
511
MOD
C
500
was
1
VF
513
refmod
OV
refvc 503
MOD
516
HD
VXINF
504
VXFIN
505
ADVX
506
VXFIN
507
HD
HD
HD
HD
gebucht
worden
ist
3
4
,
5
dann
6
PRED
NX
508
HD
ist
7
PX
515
−
HD
2
MOD
LK
514
HD
VXINF
503
HD
da
0
OV
NX
502
HD
Wenn
VC
512
ON
ADVX
501
−
ON
ADVX
509
NX
510
HD
das
8
HD
HD
nicht
9
in
10
Ordnung
11
12
KOUS
ADV
PIS
VVPP
VAPP
VAFIN
$,
ADV
VAFIN
PDS
PTKNEG
APPR
NN
−−
−−
***
−−
−−
3sis
−−
−−
3sis
nsn
−−
d
dsf
If dann is not present in the matrix clause , the wenn-clause occurs in VF. In this case,
the wenn-clause is labelled as MOD because there is no explicit correlating constituent.
It rather refers to the whole matrix clause, e.g. (Wenn da was gebucht worden ist (MOD),
ist das nicht in Ordnung.)
6.2
Questions
6.2.1
W-Questions
In general, w-questions are verb-second clauses with interrogative pronouns in VF. The
problem here is to decide on the syntactic category of the interrogative phrase.
We follow the strategy to assign PX to all PWAVs, which compositionally comprise
a preposition such as wobei, wofür, wogegen, woher, womit, woran, worauf, wovon, wozu
and also to causal PWAVs such as warum, wieso, weshalb. The (non-compositional)
PWAVs wann, wo are analysed as ADVX.
SIMPX
510
−
−
−
VF
507
V−MOD
LK
508
HD
PX
500
HD
VXFIN
501
HD
Warum
0
machen
−
MF
509
1
ON
OA
MOD
MOD
NX
502
HD
NX
503
ADVX
504
HD
ADVX
505
HD
nicht
einfach
wir
2
−
den
HD
3
Computer
4
5
VC
506
VPT
6
aus
7
?
8
PWAV
VVFIN
PPER
ART
NN
PTKNEG
ADV
PTKVZ
$.
−−
1pis
np*1
asm
asm
−−
−−
−−
−−
6.2.2
Yes - No Questions
Yes - no questions may occur in various forms, but the most typical form is the verb-first
clause:
91
SIMPX
505
−
−
LK
MF
503
504
HD
ON
OA
VXFIN
NX
NX
500
501
HD
502
−
Veruntreute
HD
die
0
HD
AWO
Spendengeld
1
2
?
3
4
VVFIN
ART
NN
NN
$.
3sit
nsf
nsf
asn
−−
Otherwise, a question mark at the end of a verb-second or verb-final clause indicates
that it is actually meant as a question:
SIMPX
508
−
−
−
MF
507
MOD
VF
504
ON
PRED
LK
505
ADJX
506
HD
NX
500
HD
Das
−
ADVX
502
ADVX
503
HD
HD
HD
ist
0
HD
VXFIN
501
doch
1
ganz
2
klar
3
?
4
5
PDS
VAFIN
ADV
ADV
ADJD
$.
nsn
3sis
−−
−−
−−
−−
SIMPX
507
−
−
−
MF
506
ON
V−MOD
PX
504
−
VC
505
HD
HD
C
500
NX
501
NX
502
−
HD
HD
Ob
Ampler
0
auf
1
HD
Sieg
2
VXFIN
503
fahre
3
?
4
5
KOUS
NE
APPR
NN
VVFIN
$.
−−
nsm
a
asm
3sks
−−
6.3
Relative Clauses
Considering relative clauses (R-SIMPX), the relative pronoun occurs in the C-field. It is
first projected to the phrase level before it is attached to the C node. The relative clause
itself is located in NF like in the following example if no other constituent follows. Its
92
edge label shows to which constituent of the matrix clause it is related. OA-MOD, for
example, suggests that the relative clause refers to OA:
SIMPX
516
−
−
−
−
−
NF
515
OA−MOD
MF
513
KOORD
500
VF
507
LK
508
ON
HD
NX
501
−
Aber
C
510
HD
−
ON
NX
504
HD
gäbe
1
−
NX
509
ADJX
503
HD
es
0
OA
−
VXFIN
502
HD
R−SIMPX
514
intelligente
Lösungen
3
,
4
OA
HD
−
die
5
VC
512
NX
505
HD
2
−
MF
511
VXFIN
506
HD
kein
6
HD
Geld
kosten
7
8
.
9
10
KON
PPER
VVFIN
ADJA
NN
$,
PRELS
PIAT
NN
VVFIN
$.
−−
nsn3
3skt
apf
apf
−−
np*
asn
asn
3pis
−−
If the head noun phrase of the relative clause is the noun phrase of a prepositional
phrase or a postmodifier within a complex phrase, the relative clause is labelled as MOD.
Additionally, there is a secondary edge label named refint (cf. 3.4.6) from the head noun
NX to the relative clause:
SIMPX
517
−
−
−
−
NF
516
MOD
R−SIMPX
515
−
MF
512
OPP
VF
506
LK
507
ON
HD
NX
500
−
Ein
HD
515
Bettenrost
0
−
mutiert
1
−
zu
2
−
NX
503
HD
einem
3
NX
510
HD
NX
502
HD
ON
PX
509
refint
−
HD
Gefängnisgitter
4
,
5
hinter
6
−
MF
514
V−MOD
PX
508
VXFIN
501
HD
−
C
513
HD
ADJX
504
VXFIN
505
HD
dem
7
VC
511
HD
HD
freier
8
Himmel
9
lockt
10
.
11
12
ART
NN
VVFIN
APPR
ART
NN
$,
APPR
PRELS
ADJA
NN
VVFIN
$.
nsm
nsm
3sis
d
dsn
dsn
−−
d
dsn
nsm
nsm
3sis
−−
The position of the relative clause in NF is justified by the fact that it does not
necessarily occur as an immediate constituent located to the right of the noun phrase
to which it refers. For example, a verb complex can occur between the noun phrase
and the relative clause (Der Bettenrost ist zu einem Gefängnisgitter mutiert, hinter dem
freier Himmel lockt.). In sentences like this, the complexity of the noun phrase (NP +
relative clause) is important. This so called heavyness follows Behaghel’s first physical
law (Behaghel 1932): complex noun phrases tend to find a position at the end of the
sentence even if they deviate from their basic order. If the relative clause does not follow
the noun phrase immediately, its unmarked position is in NF. Unless there is strong
evidence for a position in MF, the relative clause is located in NF.
93
If the relative clause and its head noun phrase are adjacent constituents in VF or MF,
the relative clause modifies the noun phrase directly as a postmodifier.
R−SIMPX
515
−
−
−
MF
514
OA
NX
513
HD
−
R−SIMPX
512
C
−
−
−
C
MF
VC
507
508
ON
NX
NX
500
−
die
die
0
HD
HD
ADVX
NX
NX
VXFIN
VXFIN
503
HD
AWO
,
1
2
511
PRED
502
HD
VC
510
ON
501
HD
509
V−MOD
wo
3
4
504
505
HD
HD
er
Kreisvorsitzender
506
HD
HD
ist
5
6
,
7
prüfte
8
9
PRELS
ART
NN
$,
PWAV
PPER
NN
VAFIN
$,
VVFIN
nsf
asf
asf
−−
−−
nsm3
nsm
3sis
−−
3sit
6.3.1
Event-modifying Relative Clauses
Relative clauses that modify an event which is not expressed by a nominal expression are
annotated as SIMPX.
SIMPX
518
−
−
−
−
NF
517
MOD
SIMPX
516
−
−
MF
514
OA
FOPP
LK
508
HD
V−MOD
PX
509
−
VXFIN
500
HD
HD
NX
501
−
NX
502
HD
HD
NX
504
HD
VC
513
HD
HD
NX
505
−
,
was
NN
VVPP
$,
PWS
APPR
PIDAT
NN
PIAT
NN
VVFIN
$.
−−
3pis
asm
asm
asn
asn
−−
−−
nsn
d
dsf
dsf
***
asm
3sis
−−
6.3.2
7
8
9
10
Szene
11
mehr
VXFIN
507
HD
HD
APPRART
6
jeder
−
NN
5
mit
NX
506
HD
ART
4
übertragen
−
haben
3
Niederdeutsche
VXINF
503
HD
OA
PX
512
VAFIN
2
ins
C
511
ON
...
1
Text
VC
510
OV
−−
0
den
−
MF
515
12
Sinn
13
macht
14
Independent Relative Clauses
Independent relative clauses (also ’nominal relative clauses’, in German ’Freie Relativsätze’) do not modify a head word but substitute an argument or adjunct in the
clause. Consequently, they are labelled SIMPX on sentential level (instead of R-SIMPX)
and they function as (sentential) subject (ON) or sentential object (OS). The latter is
not uncontroversial since they are distributed like non-sentential, nominal arguments with
respect to subcategorization restrictions.
The relative pronoun used in independent relative clauses normally belongs to the
w-class of relative pronouns such as wer or was and is tagged with the STTS tag PWS.
94
.
15
SIMPX
516
−
−
−
VF
515
ON
SIMPX
513
−
−
C
508
ON
MOD
NX
500
HD
ADVX
501
HD
Wer
MF
509
V−MOD
OV
ADJX
502
HD
VXINF
503
HD
VC
510
OV
503
VXINF
504
HD
HD
VXFIN
506
HD
NX
507
−
will
,
kommt
VAINF
VMFIN
$,
VVFIN
APPR
PPOSAT
NN
$.
ns*
−−
−−
−−
−−
3sis
−−
3sis
a
ap*
ap*
−−
SIMPX
517
−
−
3
4
5
6
−
7
auf
8
seine
HD
VVPP
2
werden
VXFIN
505
HD
PX
512
−
ADJD
1
unterhalten
LK
511
HD
HD
ADV
−
gut
refvc
PWS
0
einfach
MF
514
OPP
−
9
Kosten
10
.
11
−
NF
516
OS
SIMPX
515
−
−
VF
508
MOD
LK
509
HD
ADVX
500
HD
VXFIN
501
HD
Manchmal
NX
502
HD
C
512
OA
MF
513
ON
VC
514
HD
ADJX
503
HD
VXINF
504
HD
NX
505
HD
NX
506
HD
VXFIN
507
HD
klar
sagen
,
was
will
.
VMFIN
PIS
ADJD
VVINF
$,
PWS
PIS
VMFIN
$.
−−
3sis
ns*
−−
−−
−−
asn
ns*
3sis
−−
1
man
VC
511
OV
ADV
0
muß
ON
−
MF
510
V−MOD
2
3
4
5
man
6
7
8
9
Independent relative clauses introduced by wie are currently annotated in a different
manner. Wie is analysed as subordinating conjunction (KOUS). This type of structure
is to be revised in a subsequent release.
6.4
Coordination
Coordination is a syntactic phenomenon that occurs on the following annotation levels:
phrase level, field level, and sentence level. Within coordinations, the conjuncts are first
projected to their phrase, field, or clause level. In a second step, they are attached to
their mother node which is n-ary branching (conjunctions between the conjuncts). This
scheme is the same for all syntactic categories.
The edge labels between the mother node and the conjuncts of the coordination are
labelled as KONJ. This edge label supports the distinction between conjuncts, modifiers, and conjunctions within complex conjunctions (cf. 6.4.3), as well as the distinction
between coordinations and elliptical constructions (cf. 6.5).
In contrast to coordinating conjunctions in the KOORD-field, coordinating conjunctions in coordinations (und, oder, etc.) are directly attached to the mother node of the
conjuncts. The class of coordinating conjunctions consists of single, e.g. und, oder, aber,
als, as well as of complex conjunctions, e.g. entweder oder, weder noch, sowohl als. Generally, coordinating conjunctions may coordinate constituents of any category. Moreover,
95
they can form asymmetric coordinations in which the conjuncts belong to different syntactic categories (cf. 6.4.2).2 In order to distinguish conjunctions from conjuncts within
a coordination, their edge labels are empty.
In the following, coordination on all annotation levels as well as specific cases of
coordination, e.g. split coordinations, will be demonstrated.
6.4.1
Coordination of Phrases
Noun Phrases
NX
506
KONJ
−
KONJ
NX
504
NX
505
HD
−
NX
500
HD
NX
501
HD
−
Ende
NX
502
HD
der
0
−
NX
503
HD
Kämpfe
1
und
2
−
Verurteilung
HD
der
3
4
Selbstmandatierung
5
6
NN
ART
NN
KON
NN
ART
NN
asn
gpm
gpm
−−
asf
gsf
gsf
Prepositional Phrases
PX
504
KONJ
−
KONJ
PX
502
PX
503
−
HD
−
HD
NX
500
NX
501
HD
am
−
Arbeitsplatz
0
oder
1
in
2
HD
der
3
Familie
4
5
APPRART
NN
KON
APPR
ART
NN
dsm
dsm
−−
d
dsf
dsf
Adjectival Phrases
NX
503
−
HD
ADJX
502
KONJ
−
KONJ
ADJX
500
ADJX
501
HD
HD
Heimliche
und
0
illegale
1
Pioniertat
2
3
ADJA
KON
ADJA
NN
nsf
−−
nsf
nsf
2
If bis is used as a conjunction like in 10.000 bis (KON) 20.000 koreanischen Daewoo PKW it is
tagged as KON. But remember that von ... bis ... phrases are treated differently (cf. 4.4.1).
96
SIMPX
510
−
−
−
MF
509
PRED
VF
506
LK
507
ON
HD
NX
500
ADJX
508
VXFIN
501
HD
HD
Das
KONJ
KONJ
KONJ
KONJ
ADJX
502
ADJX
503
ADJX
504
ADJX
505
HD
klingt
0
HD
anmaßend
1
,
2
HD
pathosschwer
3
,
4
HD
elektrisch
5
,
6
laut
.
7
8
9
PDS
VVFIN
ADJD
$,
ADJD
$,
ADJD
$,
ADJD
$.
nsn
3sis
−−
−−
−−
−−
−−
−−
−−
−−
Adverbial Phrases
ADVX
502
KONJ
−
KONJ
ADVX
500
ADVX
501
HD
HD
solo
oder
0
zusammen
1
2
ADV
KON
ADV
−−
−−
−−
6.4.2
Asymmetric Coordination
Since constituents of different syntactic categories can be coordinated, it has to be decided
on a label for the mother node of the coordination. In this case, the default strategy has
been adopted to choose the syntactic category of the left-most conjunct as the category
of the entire coordination:
ADVX
504
KONJ
KONJ
ADVX
500
HD
heute
,
0
KONJ
−
KONJ
NX
501
NX
502
NX
503
HD
HD
HD
So.
1
,
2
Mo.
3
,
4
5
u.
Di.
6
7
ADV
$,
NN
$,
NN
$,
KON
NN
−−
−−
nsm
−−
nsm
−−
−−
nsm
97
SIMPX
519
−
−
−
MF
518
PRED
ADJX
517
KONJ
KONJ
KONJ
PX
515
NX
516
KONJ
VF
509
LK
510
ON
HD
NX
500
−
KONJ
VXFIN
501
HD
Die
ADJX
511
Farbpalette
0
−
KONJ
und
3
geschmackvoll
4
,
5
von
−
HD
PX
513
HD
HD
zart
2
−
ADJX
503
HD
ist
1
PX
512
ADJX
502
HD
KONJ
−
NX
514
HD
KONJ
NX
504
NX
505
HD
HD
Bordeaux
bis
8
ADVX
506
HD
Flieder
9
,
10
auch
11
−
KONJ
NX
507
NX
508
HD
HD
Orange
12
oder
13
Rosa
14
.
6
7
ART
NN
VAFIN
ADJD
KON
ADJD
$,
APPR
NN
APPR
NN
$,
ADV
NN
KON
NN
15
$.
nsf
nsf
3sis
−−
−−
−−
−−
d
dsn
a
asn
−−
−−
nsn
−−
nsn
−−
SIMPX
512
−
−
−
−
MF
511
PRED
NX
510
−
VF
505
LK
506
ON
HD
NX
500
HD
KONJ
−
VXFIN
501
PX
508
HD
VC
509
−
HD
weder
1
−
ausgesprochene
2
OV
NX
503
HD
bin
0
KONJ
ADJX
502
HD
Ich
−
NX
507
Meat−Loaf−FanIn
3
noch
4
in
5
dem
6
VXINF
504
HD
HD
Konzert
7
gewesen
8
.
9
10
PPER
VAFIN
KON
ADJA
NN
KON
APPR
ART
NN
VAPP
$.
ns*1
1sis
−−
nsf
nsf
−−
d
dsn
dsn
−−
−−
6.4.3
Coordinations with Complex Conjunctions
The conjuncts and conjunctions of a coordination with complex conjunctions are also
attached on the same level following the above mentioned rules for coordination. Both
parts of complex conjunctions like entweder oder and sowohl als are tagged as KON.
The latter one usually occurs together with the adverb auch, which is tagged as ADV,
projected to the phrase level, and then attached to the mother node of the coordination.
The same applies for nicht in coordinations with sondern. Sondern is tagged as KON,
whereas nicht is always tagged as PTKNEG:
98
16
SIMPX
516
−
−
−
FKOORD
515
KONJ
−
KONJ
MF
514
V−MOD
PRED
PX
513
−
VF
509
LK
510
MOD
HD
ADVX
500
VXFIN
501
HD
HD
Immerhin
MF
511
wird
NX
512
ON
MOD
MOD
MOD
PRED
NX
502
ADVX
503
ADVX
504
ADVX
505
ADJX
506
HD
HD
HD
HD
0
HD
es
1
nicht
2
noch
3
−
ADJX
507
HD
obendrein
5
ADJX
508
HD
kalt
4
HD
,
sondern
bei
8
HD
20
9
Grad
10
erträglich
.
6
7
ADV
VAFIN
PPER
PTKNEG
ADV
ADV
ADJD
$,
KON
APPR
CARD
NN
11
ADJD
12
$.
13
−−
3sis
nsn3
−−
−−
−−
−−
−−
−−
d
−−
dpn
−−
−−
SIMPX
517
−
−
−
MF
516
V−MOD
VF
514
OA
ADJX
515
ON
−
KONJ
−
−
KONJ
NX
512
PX
513
HD
−
−
PX
508
−
HD
NX
500
−
Papst−Besuch
in
1
HD
VXFIN
502
HD
0
NX
510
HD
NX
501
HD
Der
HD
LK
509
ADJX
503
HD
Bukarest
2
ADVX
504
HD
spielt
3
sowohl
4
NX
505
HD
außenpolitisch
als
5
6
auch
für
8
−
−
ADVX
506
HD
7
NX
511
−
HD
Rumänien
9
HD
ADJX
507
HD
selbst
10
eine
11
große
12
Rolle
.
13
14
15
ART
NN
APPR
NE
VVFIN
KON
ADJD
KON
ADV
APPR
NE
ADV
ART
ADJA
NN
$.
nsm
nsm
d
dsn
3sis
−−
−−
−−
−−
a
asn
−−
asf
asf
asf
−−
SIMPX
525
−
KONJ
−
KONJ
SIMPX
524
−
−
−
−
NF
523
OS
MF
520
ON
MOD
LK
511
HD
−
NX
501
HD
HD
haben
0
−
ADVX
512
VXFIN
500
Entweder
SIMPX
521
die
HD
OV
ADVX
502
VXINF
503
HD
HD
überhaupt
2
nicht
3
C
514
begriffen
4
MF
515
OPP
,
5
ON
PX
504
NX
505
HD
HD
worum
6
−
−
VC
516
VF
517
HD
ON
VXFIN
506
HD
geht
8
,
9
oder
10
VXFIN
508
HD
es
11
−
LK
518
HD
NX
507
HD
es
7
SIMPX
522
−
MF
519
OD
PRED
NX
509
ADJX
510
HD
ist
HD
ihnen
egal
14
.
12
13
KON
VAFIN
PDS
ADV
PTKNEG
VVPP
$,
PWAV
PPER
VVFIN
$,
KON
PPER
VAFIN
PPER
ADJD
$.
−−
3pis
np*
−−
−−
−−
−−
−−
nsn3
3sis
−−
−−
nsn3
3sis
dp*3
−−
−−
6.4.4
1
VC
513
−
15
Coordinations with Truncated Words
Truncated words are projected to the phrase level. Their edge labels are empty. The
phrases of both conjuncts are coordinated. The truncated words do not receive morphological annotation.
99
16
NX
500
KONJ
−
KONJ
NX
NX
501
502
−
HD
Bau−
und
0
Verkehrsplanungen
1
2
TRUNC
KON
NN
−−
−−
dpf
PX
503
−
HD
NX
502
KONJ
−
KONJ
NX
500
NX
501
−
bei
−
einer
0
−
SPD−
1
oder
2
HD
einer
CDU−Veranstaltung
3
4
5
APPR
ART
TRUNC
KON
ART
NN
d
dsf
−−
−−
dsf
dsf
NX
502
KONJ
−
KONJ
NX
501
−
NX
HD
ADVX
503
500
−
HD
Noch−Frauen−
und
0
bald
1
Fußballsender
2
3
TRUNC
KON
ADV
NN
−−
−−
−−
nsm
In the case of complex conjunctions, the conjuncts are annotated in the same way.
ADJX
503
−
KONJ
−
−
ADJX
ADVX
500
ADJX
501
−
sowohl
KONJ
502
HD
kultur−
0
als
1
HD
auch
2
stadtentwicklungspolitisch¦
3
4
KON
TRUNC
KON
ADV
ADJD
−−
−−
−−
−−
−−
NX
503
−
KONJ
ADVX
500
−
KONJ
NX
501
HD
−
nicht
die
0
NX
502
−
−
Sozial−
1
,
2
sondern
3
HD
die
4
Bildungsbehörde
5
6
PTKNEG
ART
TRUNC
$,
KON
ART
NN
−−
nsf
−−
−−
−−
nsf
nsf
100
Word initial TRUNCs are different from truncated words which include the second
part of a word. The latter ones are treated like complete lexical heads, because they
comprise the head morpheme of the complex word.
NX
506
HD
−
PX
505
−
HD
NX
NX
503
KONJ
504
−
KONJ
−
NX
ADJX
NX
500
501
HD
502
HD
Originaltitel
und
0
HD
HD
−fassung
"
2
8
MM
5
"
6
.
3
4
7
8
KON
NN
APPR
$(
CARD
NN
$(
$.
nsm
−−
nsf
d
−−
−−
dsm
−−
−−
6.4.5
1
von
NN
Attachment Principles of Coordination within Phrases
If two or more nominal conjuncts occur together with a common determiner and/or
adjectival phrase, first the conjuncts are projected to their phrase level and then the
determiner or the adjectival phrase is attached to the coordination on a higher level
according to the high attachment principle. Thus, the modification scope comprises the
entire coordination. The coordinated part is assigned the head function.
NX
503
−
HD
NX
502
KONJ
den
−
KONJ
NX
500
NX
501
HD
HD
Angestellten
und
0
1
Beamten
2
3
ART
NN
KON
NN
dp*
dp*
−−
dpm
NX
504
−
−
HD
NX
503
KONJ
ADJX
NX
501
HD
502
HD
türkischen
0
KONJ
NX
500
Die
−
HD
Instrumente
1
und
2
Harmonien
3
4
ART
ADJA
NN
KON
NN
np*
np*
npn
−−
npf
101
6.4.6
Coordination of Topological Fields
The conjuncts of a coordination of topological fields are either single fields (cf. 6.4.4) or a
combination of fields. Possible combinations are, for instance, (MF + VC), (LK + MF),
(LK + MF + VC). The node label for these conjuncts is FKONJ (conjunct consisting of
fields) and the mother node of a coordination of conjuncts of fields is FKOORD.
In a coordination of conjuncts of fields, the following annotation steps are involved:
1. The constituents are attached to the fields in which they occur in (MF, VC, NF,
etc.).
2. Each conjunct (concatenation of fields or single field) is labelled as FKONJ.
3. The conjuncts are attached to the general coordination field FKOORD.
SIMPX
526
−
−
FKOORD
525
KONJ
−
KONJ
FKONJ
523
FKONJ
524
−
−
−
−
−
MF
521
NF
522
OPP
OA−MOD
PX
518
MF
519
−
VF
510
HD
LK
511
ON
NX
512
HD
NX
500
−
HD
glauben
0
HD
ADJX
502
HD
Wir
LK
513
−
VXFIN
501
HD
V−MOD
an
1
die
2
totale
3
Gegenwart
4
und
5
−
ADVX
514
HD
KONJ
ADVX
504
ADVX
505
HD
HD
HD
tun
−
hier
7
KONJ
und
8
−
C
515
VXFIN
503
6
R−SIMPX
520
OA
jetzt
9
−
MF
516
OA
ON
NX
506
NX
507
NX
508
HD
HD
HD
alles
10
,
11
was
12
VC
517
HD
VXFIN
509
HD
wir
können
.
13
14
PPER
VVFIN
APPR
ART
ADJA
NN
KON
VVFIN
ADV
KON
ADV
PIS
$,
PRELS
PPER
VMFIN
$.
np*1
1pis
a
asf
asf
asf
−−
1pis
−−
−−
−−
asn
−−
asn
np*1
1pis
−−
Often, the subject of the sentence occurs only in the left field conjunct:
SIMPX
514
−
−
FKOORD
513
KONJ
−
KONJ
FKONJ
511
−
VF
506
"
MF
508
ON
MOD
HD
VXFIN
501
HD
HD
Immer
−
LK
507
ADVX
500
0
FKONJ
512
OD
VXFIN
503
HD
und
3
HD
stiehlt
4
OA
NX
504
HD
einer
2
−
MF
510
HD
NX
502
kommt
1
−
LK
509
mir
5
NX
505
−
HD
meine
6
Krise
7
"
8
9
.
10
$(
ADV
VVFIN
PIS
KON
VVFIN
PPER
PPOSAT
NN
$(
$.
−−
−−
3sis
nsm
−−
3sis
ds*1
asf
asf
−−
−−
A coordination of fields may also be an embedded structure. In this case, FKOORD
functions also as conjunct label:
102
15
16
SIMPX
521
−
−
FKOORD
520
KONJ
KONJ
FKONJ
519
−
−
FKOORD
518
KONJ
FKONJ
515
−
−
VF
508
LK
509
ON
HD
NX
500
−
Die
Älteren
1
−
VXFIN
503
HD
teurer
,
3
HD
OA
ADVX
505
HD
familiäre
5
VC
514
MOD
HD
haben
4
−
MF
513
ADJX
504
HD
2
−
NX
512
HD
ADJX
502
sind
0
LK
511
PRED
HD
KONJ
FKONJ
517
OA
MF
510
VXFIN
501
HD
−
MF
516
Verpflichtungen
6
und
7
NX
506
−
oft
8
OV
ein
9
VXINF
507
HD
HD
Haus
10
abzuzahlen
11
.
12
13
ART
NN
VAFIN
ADJD
$,
VAFIN
ADJA
NN
KON
ADV
ART
NN
VVIZU
$.
np*
np*
3pis
−−
−−
3pis
apf
apf
−−
−−
asn
asn
−−
−−
6.4.7
Attachment of Ambiguous Modifiers in Coordination
Within phrases, the modification scope of a premodifier can be ambiguous. Therefore,
high attachment is applied to preserve ambiguity. In the following example, the adverb
modifies the coordination of adjectives rather than only the first adjective:
ADJX
504
−
HD
ADJX
503
KONJ
ADVX
500
HD
−
KONJ
ADJX
501
ADJX
502
HD
Viel
HD
größer
0
und
1
brutaler
2
3
ADV
ADJD
KON
ADJD
−−
−−
−−
−−
Modifying constituents are attached to a conjunct rather than to a field if their modification scope is limited to the conjunct.
103
SIMPX
512
−
−
−
MF
511
OPP
PX
510
KONJ
VF
506
LK
507
ON
HD
−
VXFIN
501
ADVX
502
HD
HD
NX
500
HD
Wir
KONJ
PX
508
glauben
0
−
HD
−
NX
503
an
2
HD
die
3
−
HD
ADVX
504
−
nicht
1
PX
509
−
NX
505
HD
Vergangenheit
4
und
5
−
nicht
6
an
7
HD
die
8
Zukunft
9
.
10
11
PPER
VVFIN
PTKNEG
APPR
ART
NN
KON
PTKNEG
APPR
ART
NN
$.
np*1
1pis
−−
a
asf
asf
−−
−−
a
asf
asf
−−
Also in coordinations with complex conjunctions, attachment on the phrase level is
applied if possible.
SIMPX
513
−
−
−
−
VF
512
ON
NX
VC
510
511
APP
APP
OV
NX
LK
506
HD
NX
NX
500
OA
VXFIN
NX
502
−
Sprecher
0
,
1
−
Axel
2
Wallrabenstein
,
4
KONJ
KONJ
VXINF
504
HD
die
6
−
VXINF
−
wollte
5
509
−
503
HD
3
VXINF
508
HD
501
HD
Radunskis
MF
507
−
505
HD
Entscheidung
7
weder
8
HD
bestätigen
9
noch
10
dementieren
11
.
12
13
NE
NN
$,
NE
NE
$,
VMFIN
ART
NN
KON
VVINF
KON
VVINF
$.
gsm
nsm
−−
nsm
nsm
−−
3sit
asf
asf
−−
−−
−−
−−
−−
SIMPX
514
−
−
−
−
MF
513
OA
NX
512
−
VF
508
ON
Die
KONJ
−
−
KONJ
LK
509
HD
NX
500
−
−
VXFIN
501
HD
HD
ADVX
503
HD
nicht
etwa
NX
504
−
,
sondern
ADV
PPOSAT
NN
$,
KON
ADV
ART
ADJA
NN
VVPP
$.
nsf
nsf
3sit
−−
−−
asf
asf
−−
−−
−−
asn
asn
asn
−−
−−
4
5
6
7
8
gleich
9
das
10
ganze
VXINF
507
HD
PTKNEG
3
Lektüre
ADJX
506
HD
VAFIN
2
unsere
ADVX
505
HD
HD
VC
511
OV
HD
NE
1
hatte
ADVX
502
HD
−
ART
0
Lufthansa
NX
510
−
11
Flugzeug
12
rationiert
If there is more than one constituent within a conjunct, each with its own grammatical
function, these constituents are first attached to the respective field node. Then, the fields
are coordinated:
104
13
.
14
SIMPX
516
−
−
−
FKOORD
515
KONJ
−
KONJ
MF
514
V−MOD
PRED
PX
513
−
VF
509
LK
510
MOD
HD
ADVX
500
VXFIN
501
HD
HD
Immerhin
NX
512
ON
MOD
MOD
MOD
PRED
NX
502
ADVX
503
ADVX
504
ADVX
505
ADJX
506
HD
HD
HD
HD
wird
0
HD
MF
511
es
nicht
2
noch
3
HD
obendrein
5
ADJX
508
HD
kalt
4
HD
ADJX
507
,
sondern
bei
8
HD
20
9
Grad
10
erträglich
11
.
6
7
ADV
VAFIN
PPER
PTKNEG
ADV
ADV
ADJD
$,
KON
APPR
CARD
NN
ADJD
$.
−−
3sis
nsn3
−−
−−
−−
−−
−−
−−
d
−−
dpn
−−
−−
6.4.8
1
−
12
13
Coordination of Sentences
In accordance with the longest match principle, complete sentences are coordinated as
paratactic constructions when they belong to the same syntactic unit (cf. 3.4.3), i.e., they
are coordinated by a conjunction, a comma, or a dash:
SIMPX
522
KONJ
−
KONJ
SIMPX
521
−
−
−
MF
520
OPP
PX
519
−
HD
VF
NX
516
LK
509
−
−
−
−
−
NX
VF
LK
MF
VC
510
APP
APP
NX
Storchbein
1
−
setzt
2
auf
3
504
HD
die
4
HD
ADJX
503
HD
512
−
NX
502
−
Henning
0
−
VXFIN
501
−
511
HD
NX
500
Nashorn−Bürgermeister
518
HD
NX
HD
SIMPX
517
ON
Sammlung
der
6
Kräfte
8
und
9
515
MOD
OV
ADJX
VXFIN
ADVX
VXINF
506
HD
positiven
7
514
HD
505
HD
5
513
V−MOD
HD
prompt
10
wird
11
507
HD
da
12
508
HD
gepöbelt
13
.
14
15
NN
NE
NE
VVFIN
APPR
ART
NN
ART
ADJA
NN
KON
ADJD
VAFIN
ADV
VVPP
$.
nsm
nsm
nsm
3sis
a
asf
asf
gpf
gpf
gpf
−−
−−
3sis
−−
−−
−−
105
SIMPX
520
KONJ
KONJ
SIMPX
518
−
SIMPX
519
−
−
VF
514
MOD
MF
515
ON
ADVX
508
−
ADVX
500
HD
ADVX
501
HD
aber
PX
510
NX
503
HD
blieb
alles
HD
−
NX
504
HD
LK
512
HD
HD
ADVX
505
HD
NX
513
−
−
VXFIN
506
HD
ADJX
507
HD
−
nur
der
Innensenator
ist
ein
neuer
ADV
VVFIN
PIS
APPRART
NN
$(
ADV
ART
NN
VAFIN
ART
ADJA
$.
−−
−−
3sit
nsn
dsn
dsn
−−
−−
nsm
nsm
3sis
nsm
nsm
−−
1
2
3
alten
−
ADV
0
beim
−
MF
517
PRED
NX
511
−
VXFIN
502
HD
−
VF
516
ON
OPP
LK
509
HD
HD
So
−
4
5
6
7
8
9
10
11
.
12
13
A coordination may also consist of two sentences with the subject of the whole construction only occurring in the left conjunct of the coordination.
SIMPX
520
KONJ
−
KONJ
SIMPX
519
−
−
−
VF
517
SIMPX
518
OS
−
−
SIMPX
515
−
MF
516
−
−
MF
509
OA
C
500
NX
501
−
HD
Ohne
LK
511
MF
512
LK
513
ON
HD
OV
HD
HD
VXINF
502
VXINF
503
VXFIN
504
HD
sie
0
OA
VC
510
HD
gesehen
1
−
zu
2
NX
505
HD
haben
3
?
4
,
5
−
kontert
der
6
7
PX
514
−
VXFIN
506
HD
HD
Popstar
8
und
9
OPP
−
wirft
ADVX
508
HD
einen
10
HD
NX
507
11
HD
Blick
12
nach
13
rechts
14
.
15
16
KOUS
PPER
VVPP
PTKZU
VAINF
$.
$,
VVFIN
ART
NN
KON
VVFIN
ART
NN
APPR
ADV
$.
−−
ap*3
−−
−−
−−
−−
−−
3sis
nsm
nsm
−−
3sis
asm
asm
d
−−
−−
Subclauses (either in VF or in NF) with or even without a conjunction can also be
coordinated.
SIMPX
525
−
−
−
MF
524
MOD
MOD
ON
SIMPX
523
KONJ
−
KONJ
SIMPX
521
−
VF
514
−
LK
515
PRED
SIMPX
522
−
MF
516
HD
ON
OV
refvc
ADJX
500
HD
Unklar
VXFIN
501
ADVX
502
ADVX
503
HD
HD
HD
ist
0
aber
1
C
504
−
noch
2
,
3
NX
505
−
wie
HD
diese
−
VC
517
Leistung
6
OV
506
HD
VXINF
506
VXINF
507
VXFIN
508
HD
HD
HD
beurteilt
7
werden
8
−
C
518
ON
kann
−
10
und
11
OPP
NX
509
PX
510
HD
HD
wer
12
PRED
ADJX
511
HD
dafür
13
VC
520
zuständig
14
OV
HD
VXINF
512
VXFIN
513
HD
HD
sein
15
soll
16
.
4
5
ADJD
VAFIN
ADV
ADV
$,
KOUS
PDAT
NN
VVPP
VAINF
VMFIN
$(
KON
PWS
PROP
ADJD
VAINF
VMFIN
$.
−−
3sis
−−
−−
−−
−−
nsf
nsf
−−
−−
3sis
−−
−−
ns*
−−
−−
−−
3sis
−−
106
9
−
MF
519
17
18
6.4.9
Paratactic Constructions with denn and weil
Paratactic constructions consisting of verb-second clauses conjoined by the conjunctions
denn and weil, which also occur in the PARORD-field in the beginning of a sentence,
are treated as equal conjuncts (verb-second instead of verb-final in weil-clause). In order
to distinguish coordination of sentences with conjunct of the PARORD field from the
above mentioned coordinations of sentences, these paratactic constructions are labelled
as P-SIMPX instead of SIMPX.
P−SIMPX
519
KONJ
−
KONJ
SIMPX
SIMPX
517
−
518
−
−
−
−
−
VF
MF
515
516
ON
OA
NX
LK
508
KONJ
NX
500
HD
LK
511
NX
512
HD
OV
ON
HD
NX
VXFIN
VXINF
NX
VXFIN
502
HD
und
VF
510
KONJ
501
0
VC
509
−
Kilos
−
503
HD
Fitneß
1
504
HD
sollen
3
,
4
denn
5
514
HD
OV
NX
VXINF
506
HD
Wesemann
6
−
505
HD
stimmen
2
VC
513
−
will
7
die
8
−
Tour
9
de
10
507
−
HD
France
11
gewinnen
12
.
13
14
NN
KON
NN
VMFIN
VVINF
$,
KON
NE
VMFIN
ART
NE
NE
NE
VVINF
$.
npn
−−
nsf
3pis
−−
−−
−−
nsm
3sis
asf
asf
asf
asf
−−
−−
6.4.10
Conjunctions Occurring with Isolated Phrases
If a conjunct occurs isolated with a conjunction, high attachment is applied like in complete coordinations. But for isolated conjuncts, the conjunct is annotated as the head of
the construction (HD instead of KONJ).
ADVX
501
−
HD
ADVX
500
HD
und
jetzt
0
1
KON
ADV
−−
−−
ADVX
502
−
Oder
−
HD
ADVX
500
ADVX
501
HD
HD
eben
0
nicht
1
.
2
3
KON
ADV
PTKNEG
$.
−−
−−
−−
−−
107
If there are modifiers which do not modify the conjunct itself because they are ambiguous or might modify something else rather than the conjunct, they are attached on
the same (high) level as the conjunction:
NX
505
−
HD
−
−
−
PX
504
−
NX
500
HD
Und
das
0
ADVX
501
ADVX
502
HD
HD
auch
1
HD
NX
503
HD
noch
2
ohne
3
Mehrvergütung
4
.
5
6
KON
PDS
ADV
ADV
APPR
NN
$.
−−
nsn
−−
−−
a
asf
−−
NX
506
−
−
−
HD
NX
505
HD
−
PX
504
−
PX
500
ADVX
501
HD
und
NX
502
HD
damit
0
−
auch
1
HD
die
2
NX
503
HD
HD
Nervosität
3
im
4
Nato−Hauptquartier
5
.
6
7
KON
PROP
ADV
ART
NN
APPRART
NN
$.
−−
−−
−−
nsf
nsf
dsn
dsn
−−
6.4.11
Split Coordinations
Closely related to isolated conjuncts are split coordinations. Generally, the left conjunct
of a split coordination is located in MF, in rare cases in VF, and the right conjunct occurs
in NF. In order to express the relation between them, the left conjunct carries the label of
its grammatical function (ON, OA, OD, etc.) whereas the right conjunct carries a label
that denotes that it is the conjunct of this grammatical function (e.g. ONK, OAK, ODK,
etc.). In asymmetric coordination, the syntactic category of the second split conjunct
determines the syntactic category one level higher up:
108
SIMPX
513
−
−
−
−
MF
511
NF
512
OA
VF
507
LK
508
ON
HD
NX
500
−
Jedes
PX
509
HD
Ja−Wort
0
OAK
NX
510
−
VXFIN
501
HD
OPP
zieht
1
HD
KONJ
KONJ
KONJ
NX
502
NX
503
NX
504
NX
505
NX
506
HD
HD
HD
HD
HD
Applaus
nach
2
3
sich
,
4
5
Unterschriften
6
,
7
Küsse
8
,
9
Händeschütteln
.
10
11
12
PIDAT
NN
VVFIN
NN
APPR
PRF
$,
NN
$,
NN
$,
NN
$.
nsn
nsn
3sis
asm
d
ds*3
−−
apf
−−
apm
−−
asn
−−
SIMPX
514
−
−
−
−
−
MF
NF
512
513
ON
VF
LK
507
−
ADJX
VXFIN
ADVX
501
hat
0
nicht
1
NX
503
−
HD
die
4
KONJ
NX
505
−
Jöns
3
−
VXINF
504
−
Karin
2
511
OV
NX
HD
NX
510
KONJ
502
HD
Selbstverständlich
VC
509
HD
HD
ONK
NX
508
MOD
500
OA
Flächen
5
506
HD
−
gebucht
,
6
7
sondern
8
HD
die
9
SPD
10
.
11
12
ADJD
VAFIN
PTKNEG
NE
NE
ART
NN
VVPP
$,
KON
ART
NE
$.
−−
3sis
−−
nsf
nsf
apf
apf
−−
−−
−−
nsf
nsf
−−
SIMPX
511
−
−
−
−
LK
509
NF
510
HD
VF
505
ONK
VXFIN
506
KONJ
NX
500
VXFIN
501
VXFIN
502
HD
HD
HD
Lausbuben
−
MF
507
ON
sind
0
KONJ
und
1
PRED
KONJ
NX
503
ADJX
504
HD
bleiben
2
ADJX
508
−
HD
sie
3
und
4
unwiderstehlich
5
.
6
7
NN
VAFIN
KON
VVFIN
PPER
KON
ADJD
$.
npm
3pis
−−
3pis
np*3
−−
−−
−−
6.5
Elliptical Constructions
In elliptical constructions, syntactically necessary linguistic elements are missing which
can be reconstructed from the context or the speech situation. Elliptical constructions
appear on the phrase level as well as on the sentence level.
The model of topological fields does not make any assumptions about dependency
relations, but it allows that topological fields may be left empty. For the description of
elliptical sentence constructions, the scheme of topological fields is an appropriate model
because neither crossing branches nor traces have to be used to annotate the surface
structure of a sentence.
109
In elliptical phrases, the head word is missing. They are annotated like phrases
without a head. Therefore, the edge labels of an elliptical phrase are empty:
SIMPX
510
−
−
−
VF
508
MF
509
ON
OD
NX
504
−
LK
505
−
HD
−
irischen
Hinweisschilder
2
den
3
HD
ADVX
503
HD
seien
1
−
ADJX
502
HD
0
ADJX
507
−
VXFIN
501
HD
die
NX
506
HD
ADJX
500
PRED
HD
walisischen
4
ziemlich
ähnlich
5
6
7
ART
ADJA
NN
VAFIN
ART
ADJA
ADV
ADJD
npn
npn
npn
3pks
dpn
dpn
−−
−−
KONJ
−
PX
503
KONJ
PX
502
−
HD
PX
NX
500
501
−
HD
in
und
0
um
1
Berlin
2
3
APPR
KON
APPR
NE
d
−−
a
asn
PX
505
−
HD
NX
504
HD
−
NX
503
−
−
ADJX
500
HD
vom
15.
ADJX
501
HD
NX
502
HD
4.
99
APPRART
ADJA
ADJA
CARD
dsm
dsm
dsm
−−
0
1
2
3
SIMPX
521
−
−
−
FKOORD
520
KONJ
KONJ
MF
517
V−MOD
VF
509
LK
510
ON
HD
NX
500
−
Die
FDP
0
in
2
,
6
in
7
−
NX
507
und
10
in
11
HD
Brandenburg
12
HD
ADJX
508
HD
3.300
9
NX
516
HD
HD
Sachsen
8
−
ADJX
506
HD
OA
PX
515
−
NX
505
4.000
5
MF
519
V−MOD
NX
514
HD
HD
knapp
4
−
ADJX
504
HD
Thüringen
3
KONJ
OA
PX
513
−
ADVX
503
HD
hat
1
−
NX
502
HD
V−MOD
NX
512
HD
VXFIN
501
HD
OA
PX
511
−
−
MF
518
2.000
13
Mitglieder
14
15
ART
NE
VAFIN
APPR
NE
ADV
CARD
$,
APPR
NE
CARD
KON
APPR
NE
CARD
NN
nsf
nsf
3sis
d
dsn
−−
−−
−−
d
dsn
−−
−−
d
dsn
−−
apn
110
SIMPX
517
−
−
FKOORD
516
−
KONJ
KONJ
FKONJ
514
FKONJ
515
−
ON
C
500
−
Ob
NX
501
−
−
MF
510
MOD
HD
−
VC
511
OA
ADVX
502
HD
−
nun
weniger
OV
NX
503
VXINF
504
HD
HD
V−MOD
VXINF
505
HD
ADJX
506
HD
ADV
PIAT
NN
VVINF
VVINF
KON
ADJD
NN
VVINF
VMFIN
−−
nsm
nsm
−−
***
apf
−−
−−
−−
−−
apf
−−
3sis
5
6
7
8
Parkgebühren
HD
VXFIN
509
HD
NN
4
flächendeckend
OV
VXINF
508
HD
Senat
3
oder
NX
507
HD
der
2
lassen
OA
ART
1
reinigen
OV
refvc 504
VC
513
KOUS
0
Straßen
−
MF
512
9
10
kassieren
11
will
12
In elliptical sentence constructions, specific topological fields are not occupied. All
constituents are attached to the appropriate field. In the first example, LK in the second
conjunct is missing. In the second example, the subject is in NF and the main clause is
lacking a verbal constituent:
SIMPX
512
−
−
LK
506
ON
VXFIN
501
HD
HD
Fall
0
−
VF
508
MF
509
PRED
ON
PRED
ADJX
502
NX
503
ADJX
504
HD
ist
1
−
MF
507
HD
NX
500
Der
KONJ
SIMPX
511
−
VF
505
−
KONJ
SIMPX
510
−
brisant
2
,
3
HD
die
HD
Mischung
4
5
explosiv
6
.
7
8
ART
NN
VAFIN
ADJD
$,
ART
NN
ADJD
$.
nsm
nsm
3sis
−−
−−
nsf
nsf
−−
−−
SIMPX
513
−
−
NF
512
ON
SIMPX
511
−
−
−
MF
510
OA
ON
MF
507
PRED
APP
ADJX
500
HD
Fein
,
0
MOD
NX
508
C
501
NX
502
−
HD
daß
sich
APP
NX
503
−
HD
NX
504
HD
die
3
VC
509
HD
Achse
4
ADJX
505
HD
Hamburg−Köln
5
VXFIN
506
HD
langsam
6
festigt
2
ADJD
$,
KOUS
PRF
ART
NN
NE
ADJD
VVFIN
$.
−−
−−
−−
as*3
nsf
nsf
nsn
−−
3sis
−−
111
7
.
1
8
9
Chapter 7
The Annotation of Specific Syntactic
Phenomena
7.1
Superlative and Comparative Forms
7.1.1
Superlative Forms
The particle am, which occurs as a particle with an adjective or an adverb in superlative constructions, is tagged as PTKA. Both, the particle and the adjective/adverb are
attached on the same level forming an adverbial/adjectival phrase:
SIMPX
512
−
−
−
VF
MF
510
511
OA
ON
NX
FOPP
PX
507
−
HD
NX
501
HD
HD
vorigen
0
Sonntag
1
ADJX
502
−
hätte
2
509
−
VXFIN
500
VC
508
HD
ADJX
Den
MOD
LK
506
−
−
−
Frank
3
503
−
Michael
4
−
Nehr
5
OV
NX
VXINF
504
HD
am
6
HD
−
liebsten
7
aus
8
dem
9
505
HD
HD
Kalender
10
gestrichen
11
.
12
13
ART
ADJA
NN
VAFIN
NE
NE
NE
PTKA
ADJD
APPR
ART
NN
VVPP
$.
asm
asm
asm
3skt
nsm
nsm
nsm
−−
−−
d
dsm
dsm
−−
−−
7.1.2
The Comparative Particles wie and als
Comparative particles in German are als and wie, in rare cases also denn (e.g. Die werden
dort seliger schlummern denn je.). These particles are tagged as KOKOM and occur with
all types of syntactic phrases (NX, ADVX, PX, etc.). They are directly attached to an
adjacent comparative phrase. In case of a comparative phrase with a postmodifier, they
are directly attached to the highest node of the complex phrase.
A comparative phrase can occur as an adjacent postmodifier of the head phrase:
112
SIMPX
516
−
−
−
VF
515
MOD
SIMPX
513
MF
514
−
−
ON
PRED
MF
511
ADJX
512
V−MOD
HD
ADJX
506
−
VC
507
HD
ADJX
500
HD
HD
VXINF
501
VXFIN
502
HD
HD
HD
Rein
musikalisch
LK
508
gesehen
0
1
NX
510
−
NX
503
−
ist
2
−
ADJX
509
HD
das
3
HD
−
−
−
ADJX
504
ADJX
505
HD
Album
HD
wesentlich
4
schlanker
5
6
als
7
das
8
erste
9
.
10
11
ADJD
ADJD
VVPP
VAFIN
ART
NN
ADJD
ADJD
KOKOM
ART
ADJA
$.
−−
−−
−−
3sis
nsn
nsn
−−
−−
−−
nsn
nsn
−−
SIMPX
510
−
−
−
MF
509
ON
OA
NX
508
HD
−
NX
505
−
C
500
NX
501
−
HD
daß
HD
−
als
HD
HD
ADJX
503
HD
1
VC
507
−
ADJX
502
sie
0
NX
506
−
VXFIN
504
HD
ehrenamtliche
2
Vorsitzende
ein
3
4
HD
dienstliches
5
Handy
6
hat
7
8
KOUS
PPER
KOKOM
ADJA
NN
ART
ADJA
NN
VAFIN
−−
nsf3
−−
nsf
nsf
asn
asn
asn
3sis
If there is a long-distance dependency between the comparative phrase and the head
phrase, the dependency relation is denoted with the respective X-MOD label.
R−SIMPX
511
−
−
−
−
MF
510
OA
C
505
ON
−
NX
500
HD
V−MOD
NX
506
PX
507
HD
−
HD
ADVX
501
mehr
1
nach
2
NX
504
HD
Bremerhaven
3
OA−MOD
VXFIN
503
HD
fünfmal
0
NF
509
HD
NX
502
HD
der
VC
508
−
liefert
4
HD
als
5
Daewoo
6
7
PRELS
ADV
PIS
APPR
NE
VVFIN
KOKOM
NE
nsm
−−
***
d
dsn
3sis
−−
ns*
In case of a long-distance dependency between the comparative phrase and the main
verb (cf. 4.7.9), the comparative phrase is either a complement (e.g. PRED) or an ambiguous or unambiguous modifier of the main verb (MOD or V-MOD).
113
SIMPX
510
−
−
−
VF
509
V−MOD
PX
508
−
HD
NX
LK
505
APP
HD
ON
PRED
NX
NX
VXFIN
NX
NX
501
−
HD
dem
502
HD
Motto
0
507
APP
500
Unter
MF
506
"
1
2
503
HD
Kino−Extrem
3
"
4
504
−
agiert
5
HD
der
6
−
Regisseur
7
HD
als
8
Filmjockey
9
10
APPR
ART
NN
$(
NN
$(
VVFIN
ART
NN
KOKOM
NN
d
dsn
dsn
−−
dsn
−−
3sis
nsm
nsm
−−
nsm
SIMPX
509
−
−
−
VF
508
MOD
PX
507
−
−
HD
NX
504
−
LK
505
−
HD
HD
ADJX
500
VXFIN
501
HD
Wie
in
0
den
1
MF
506
HD
meisten
2
Musicals
3
PRED
NX
502
ADJX
503
−
ist
4
ON
HD
die
5
HD
Handlung
6
simpel
7
8
KOKOM
APPR
ART
PIDAT
NN
VAFIN
ART
NN
ADJD
−−
d
dpn
dpn
dpn
3sis
nsf
nsf
−−
SIMPX
516
−
−
−
−
MF
515
ON
MOD
VF
512
V−MOD
SIMPX
513
−
PX
507
−
LK
508
HD
HD
NX
500
HD
NX
509
−
−
VXFIN
501
HD
−
VC
510
HD
HD
ADJX
502
HD
C
503
−
VXINF
504
HD
KONJ
VXINF
505
HD
VXINF
506
HD
Arsten
die
neue
,
wie
,
nachgearbeitet
NE
VAFIN
ART
ADJA
NN
$,
KOUS
VVPP
$,
VVPP
KON
VVPP
$.
d
dsn
3sis
nsf
nsf
nsf
−−
−−
−−
−−
−−
−−
−−
−−
2
3
4
5
6
7
geplant
VXINF
511
−
In
1
Strecke
KONJ
APPR
0
wird
VC
514
OV
8
9
10
und
11
begrünt
12
.
13
The high attachment principle applies when the comparative particle has scope over
a coordination of phrases (cf. 6.4.5). In this case, the two conjuncts are coordinated first.
Then the particle is attached on a higher level.
114
NX
503
−
HD
NX
502
KONJ
−
KONJ
NX
NX
500
−
wie
Pete
0
501
−
−
Sampras
1
oder
2
−
Yewgeni
3
Kafelnikov
4
5
KOKOM
NE
NE
KON
NE
NE
−−
dsm
dsm
−−
dsm
dsm
7.2
Verbal and Adjectival Use of Participles
In German, verbal participles which are passive verb forms (Der Mensch wird angesehen)
can be used as adjectives: it can either function as an attribute adjective (der angesehene
Mensch) or - depending on the context - also as a predicative adjective (der Mensch ist
angesehen.). In contrast to the auxiliary werden in verbal passives, the auxiliary sein
is used in constructions with adjectival passives. Concerning the problematic distinction
between verbal and adjectival passives, we adapted the criteria in the Stuttgart-Tübingen
tag set (STTS) (Schiller et al. 1995).1
1. Can the sentence be transformed into active form keeping the same semantics? If
yes → VVPP
2. Is there a von-PP or an equivalent PP that gives evidence for verb semantics? If
yes → VVPP
3. Is it possible to substitute the word in questions by a semantically similar adjective?
If yes → ADJD
The following two tree structures show the annotation of the verbal and adjectival
passives of the verbal participle angesehen. In the first example, the verbal participle is
analysed as a VVPP in VC. In the second example, the verbal participle has an adjectival
reading and is annotated as an ADJD in MF.
1
Concerning the differences between verbal and adjectival passives in English cf. Bresnan (1995).
115
SIMPX
509
−
−
−
MF
508
MOD
ON
−
C
500
−
daß
ADVX
501
HD
VC
507
HD
ADJX
502
HD
ADJX
503
−
ein
anderes
ADJA
NN
KOKOM
ADJD
VVPP
VAFIN
−−
−−
nsn
nsn
nsn
−−
−−
−−
3sit
−
−
3
4
5
gültig
VXFIN
505
HD
ART
2
als
HD
VXINF
504
HD
ADV
1
Argument
HD
OV
KOUS
0
auch
PRED
NX
506
−
6
angesehen
wurde
7
8
SIMPX
517
−
−
NF
516
ON
SIMPX
515
−
−
−
MF
514
ON
PRED
ADJX
513
−
VF
507
ON−MOD
NX
500
HD
LK
508
HD
MF
509
PRED
VXFIN
501
HD
ADJX
502
HD
NX
510
−
C
503
−
HD
−
ADJX
504
HD
VC
512
HD
HD
ADVX
505
HD
Es
ist
schade
,
daß
so
wenig
VAFIN
ADJD
$,
KOUS
ADJA
NN
ADV
ADV
ADJD
VAFIN
nsn3
3sis
−−
−−
−−
npf
npf
−−
−−
−−
3pis
7.3
1
2
3
4
5
Leistungen
VXFIN
506
HD
PPER
0
akademische
HD
ADVX
511
6
7
8
angesehen
9
sind
10
Topicalization
Topicalization is almost exclusively found in verb-second clauses. Consequently, the subject is not in the first position of the clause. Topicalized constructions bring about word
order phenomena which differ from those occurring in MF, e.g., non-finite parts of VC
are not allowed in MF.
Our annotation principles demand to analyse the topicalized verb complex and its
non-finite parts as VC in the first position of the clause. VC is then attached to VF. If a
part of MF is topicalized along with VC, first MF and VC are combined to form FKONJ
before they are attached to VF:
116
SIMPX
509
−
−
−
VF
507
MF
508
−
ON
VC
504
LK
505
OV
HD
VXINF
500
VXFIN
501
HD
HD
Geplant
V−MOD
PX
506
−
NX
502
NX
503
−
war
0
HD
HD
der
1
HD
Papst−Besuch
seit
2
3
langem
.
4
5
6
VVPP
VAFIN
ART
NN
APPR
NN
$.
−−
3sit
nsm
nsm
d
dsn
−−
SIMPX
517
−
−
−
VF
515
MF
516
−
ON
V−MOD
FKONJ
513
PX
514
−
−
−
HD
MF
511
FOPP
NX
512
PRED
HD
PX
507
−
HD
NX
500
ADJX
501
HD
Auf
HD
Pioneer
0
aufmerksam
1
VC
508
LK
509
OV
HD
VXINF
502
VXFIN
503
HD
HD
geworden
2
NX
510
−
NX
504
−
war
3
−
HD
der
4
NX
505
HD
BUND
5
durch
6
HD
Informationen
7
HD
ADJX
506
französischer
8
Bauern
9
.
10
11
APPR
NE
ADJD
VAPP
VAFIN
ART
NE
APPR
NN
ADJA
NN
$.
a
asm
−−
−−
3sit
nsm
nsm
a
apf
gpm
gpm
−−
7.4
Headlines
The syntax of headlines differs from other syntactic constructions in so far as headlines2
often lack the finite verb or a verb at all. If a headline has only an infinitive, the case
assignment follows the preference principle formulated in 5.2. Therefore, we assume in
general the more plausible grammatical function in each case: a passive constructions
with ON in MF if the verb in VC is a past participle and an active construction with OA
in MF if the verb in VC is an infinitive.
2
The identifier “HEADLINE” is inserted into the comment line above the sentence for each syntactic
unit which is marked as a headline in the original data.
117
SIMPX
507
−
−
MF
506
ON
V−MOD
NX
503
PX
504
−
HD
−
VC
505
HD
ADJX
500
HD
NX
501
HD
VXINF
502
HD
20
Dissidenten
0
in
1
HD
China
2
festgenommen
3
4
CARD
NN
APPR
NE
VVPP
−−
npm
d
dsn
−−
SIMPX
504
−
−
MF
502
VC
503
OA
HD
NX
500
VXINF
501
HD
HD
WBM−Chefs
ablösen
0
1
NN
VVINF
apm
−−
A headline can also consist of an elliptical sentence (cf. 6.5):
SIMPX
506
−
−
MF
505
PRED
VF
ADJX
507
503
ON
−
NX
ADJX
500
HD
501
HD
HD
Welthandelsorganisationen¦
vollkommen
0
kopflos
1
2
NN
ADJD
ADJD
nsf
−−
−−
Headlines can also consist of more than one syntactic structure, for instance, separated
by a colon or a dash (cf. 4.7.2 and 5.2):
118
SIMPX
505
−
ON
HD
NX
500
NX
501
HD
:
0
7.5
LK
504
HD
Rechtschreibreform
−
VF
503
VXFIN
502
HD
Gegner
1
klagen
2
3
NN
$.
NN
VVFIN
nsf
−−
npm
3pis
Discourse Markers
Generally, discourse markers are expressions or phrases of greeting, apologizing, thanking,
short emotional utterances, and interjections. Their node label is DM. The edge label of
a discourse marker is empty, i.e., it does not have a head. Typical discourse markers are:
ja, nein, hallo, oh, aha, pst, nunja, gewiß, toll, nun ja, etc.
In most cases, discourse markers occur as isolated expressions. Interjections, tagged
as ITJ, are directly projected to DM without internal structure. The same applies for
answer particles (PTKANT):
DM
500
−
Oh
0
ITJ
−−
DM
500
−
ja
0
PTKANT
−−
Phrases which function as discourse markers are first projected to their phrase level
before they are assigned the node label DM.
DM
501
−
ADVX
500
HD
gewiß
0
ADV
−−
119
DM
501
−
NX
500
−
HD
Keine
Ahnung
0
1
PIAT
NN
asf
asf
DM
502
−
NX
501
−
HD
ADJX
500
HD
Liebe
tazzen
0
1
ADJA
NN
np*
np*
Isolated conjunctions and foreign language discourse markers are tagged according to
their part of speech (KON and FM) and are projected to DM:
DM
500
−
Und
0
KON
−−
DM
500
−
pardon
0
FM
−−
Discourse markers may also consist of an interjection or an answer particle and a
phrase:
DM
501
−
−
ADVX
500
HD
Nun
ja
0
.
1
2
ADV
PTKANT
$.
−−
−−
−−
In some cases, discourse markers have a grammatical function within a phrase or a
clause. Therefore, they are attached to the syntactic structure:
120
PX
503
−
HD
NX
502
APP
APP
NX
500
DM
501
−
mit
HD
den
−
Worten
0
1
"
2
−
aha
3
,
4
−
aha
5
,
6
aha
7
8
APPR
ART
NN
$(
ITJ
$,
ITJ
$,
ITJ
d
dpn
dpn
−−
−−
−−
−−
−−
−−
NX
502
HD
−
NCX
DM
500
501
HD
−
Welt
(
0
−
Oh
1
,
2
no
3
!
4
)
5
6
NN
$(
ITJ
$,
FM
$.
$(
dsf
−−
−−
−−
−−
−−
−−
7.6
Parentheses
Parentheses occur as interjective utterances within a sentence. Since there is no dependency relation between the parenthesis and the rest of the construction, the parenthesis is
not attached to the surrounding constituents. Often parentheses occur as SIMPX-clauses.
Insertions like sagte Mehmet Scholl into direct speech are also annotated as parenthesis.3
SIMPX
515
−
−
−
−
VF
514
ON
NX
513
HD
−
NX
508
−
−
NX
500
−
HD
HD
Vielzahl
0
der
1
HD
betroffenen
2
−
VXFIN
502
Mieter
3
PX
503
HD
daher
5
,
6
VC
512
MOD
HD
ADVX
504
HD
sei
4
MF
511
NX
510
V−MOD
HD
ADJX
501
HD
Eine
LK
509
OV
NX
505
HD
so
Bertermann
,
VXINF
507
HD
HD
bereits
10
ausgezogen
11
.
7
8
ART
NN
ART
ADJA
NN
VAFIN
PROP
$,
ADV
NE
$,
ADV
VVPP
$.
nsf
nsf
gpm
gpm
gpm
3sks
−−
−−
−−
nsm
−−
−−
−−
−−
3
9
ADVX
506
12
On the TüBa-D/Z web page (http://www.sfs.uni-tuebingen.de/en/tuebadz.shtml), the treebank is also available in the Penn Treebank format. For this format, parentheses are attached to the tree
structure with the edge label PAR. For further details about the Penn Treebank format cf. 9.
121
13
−
−
VF
509
ON
ON
NCX
501
−
HD
HD
Kuratorium
0
,
1
das
2
−
−
VF
510
NCX
500
Ein
SIMPX
516
SIMPX
515
LK
511
MF
512
HD
MOD
VXFIN
502
ADVX
503
HD
HD
ist
3
−
−
PRED
−
der
OA
VXFIN
505
HD
5
MF
514
HD
NCX
504
wohl
4
LK
513
NCX
506
HD
Gedanke
6
,
7
MOD
PRED
ADVX
507
ADJX
508
HD
macht
8
HD
sich
9
HD
immer
10
gut
11
.
12
13
ART
NN
$,
PDS
VAFIN
ADV
ART
NN
$,
VVFIN
PRF
ADV
ADJD
$.
nsn
nsn
−−
nsn
3sis
−−
nsm
nsm
−−
3sis
as*3
−−
−−
−−
SIMPX
512
SIMPX
511
−
−
−
VF
LK
506
MF
507
LK
508
MF
509
510
PRED
HD
ON
HD
ON
MOD
ADJX
VXFIN
NX
VXFIN
NX
ADVX
500
501
HD
"
−
−
HD
Schön
0
502
"
1
,
−
sagte
−
Mehmet
4
503
HD
Scholl
5
,
6
"
504
HD
ist
das
nicht
10
"
11
12
.
2
3
7
8
$(
ADJD
$(
$,
VVFIN
NE
NE
$,
$(
VAFIN
PDS
PTKNEG
$(
$.
−−
−−
−−
−−
3sit
nsm
nsm
−−
−−
3sis
nsn
−−
−−
−−
122
9
505
HD
13
Chapter 8
Criteria for the Distinction of
Grammatical Functions
8.1
Subcategorization of Verbs
The TüBa-D/Z-Verblist document1 lists all verbs occurring in the treebank with their
specific subcategorization frames. This reference list guarantees the consistent annotation
of grammatical functions. For a detailed description of constructing the verb list see
(Hinrichs and Telljohann 2009).
8.2
Subcategorization of PREDs
Since constituents which predicates subcategorize for have grammatical function within
a sentence, they are neither marked as PRED-MOD nor attached to the predicate itself. These constituents are attached to a field and assigned the respective grammatical
function like the constituent which is marked as FOPP in the following examples:
1
In case of interest, please refer to web page (http://www.sfs.uni-tuebingen.de/en/tuebadz.
shtml) for contact information.
123
SIMPX
512
−
−
−
VF
511
FOPP
PX
509
MF
510
−
HD
ON
NX
506
LK
507
HD
−
NX
500
−
Für
HD
den
0
NX
501
−
Erfolg
1
HD
des
2
Volksbegehren
3
NX
508
HD
−
VXFIN
502
ADVX
503
HD
HD
sind
4
PRED
−
HD
ADJX
504
ADJX
505
HD
etwa
5
HD
243.000
6
Unterschriften
7
erforderlich
.
8
9
10
APPR
ART
NN
ART
NN
VAFIN
ADV
CARD
NN
ADJD
$.
a
asm
asm
gsn
gsn
3pis
−−
−−
npf
−−
−−
SIMPX
511
−
−
−
VF
509
MF
510
ON
OA
NX
505
−
LK
506
−
HD
Brüder
1
−
sich
3
−
NX
503
HD
fühlen
2
HD
NX
502
HD
älteren
0
ADJX
508
−
VXFIN
501
HD
PRED
PX
507
HD
ADJX
500
Die
OPP
für
4
ADVX
504
HD
ihre
5
HD
HD
Schwestern
6
sehr
7
verantwortlich
8
.
9
10
ART
ADJA
NN
VVFIN
PRF
APPR
PPOSAT
NN
ADV
ADJD
$.
npm
npm
npm
3pis
ap*3
a
apf
apf
−−
−−
−−
8.3
Distinction of FOPP, OPP, and V-MOD
One of the major problems is to distinguish, whether a given PP is an obligatory (OPP)
or an optional (FOPP) complement of a specific verb in a specific reading, or whether it
is a free adjunct (V-MOD) of that verb.
The TüBa-D/Z-Verblist is intended as a reference for these problematic cases.
In the following, we will briefly describe what criteria have been used in order to
decide about the subcategorization with respect to PP complements/modifiers:
1. A PP is called OPP within a sentence if the sentence were ungrammatical without
the OPP (or if there was at least a very noticeable change of meaning). For instance,
Sie gehen [OPP gegen die Faschisten] vor./ Das Gesetz ist [OPP in Kraft] getreten.
2. A PP is called FOPP if it can be left out of this specific sentence without causing
ungrammaticality (or a very noticeable change of meaning) and if its preposition is
selected by this specific verb. For instance, Insgesamt berichtet die Polizei [FOPP
von 19 Festnahmen und 98 Ingewahrsamnahmen]./Später würden wir [FOPP über
124
Auswandern] nachdenken. Here, the prepositions select these specific verbs and the
PPs cannot be added to any arbitrary verb (which is possible for free adjuncts).
In addition, in passive clauses, the subject of the original active clause, which has
the form of a prepositional phrase, is marked as FOPP (Sie wurden [FOPP von
Autonomen] umringt.).
3. A PP is called V-MOD if its preposition is not selected by this specific verb, i.e., it
can be exchanged by any other modifying PP, and similarly, this PP can occur with
arbitrary verbs (Nur [V-MOD im griechischen Lager] gab es Probleme). Typical
V-MODs are temporal or local adjuncts specifying time and location of the action,
event, or state expressed by the verb.
8.4
Distinction of MOD, MOD-MOD, and V-MOD
A typical case of modification of modifiers is a temporal expression (V-MOD) that further
specifies another temporal expression (MOD-MOD) in the same clause:
1. [V-MOD am Samstag] finden [MOD-MOD ab 16 Uhr] Führungen statt.
[MOD-MOD Wann] finden [V-MOD am Samstag] Führungen statt?
2. [MOD da] finden [V-MOD am Samstag] Führungen statt.
[V-MOD wann] finden [MOD da] Führungen statt?
[MOD dann] finden [V-MOD am Samstag] Führungen statt.
da, dann, etc. can be either temporal, causal, consequential, or local expressions.
Thus, one cannot make sure whether the following time expression am Samstag
really refers to them. The only obvious observation is that the time expression is a
V-MOD in any case.
For resumptive constructions (LV), there is also a clear criterion concerning the modification relations. Within a verb-second clause, a modifier occurring in VF is MOD/XMOD, whereas the modifier in LV is MOD-MOD, not vice versa, because the modifier in
VF occurs within the core of the sentence, whereas the modifier in LV has to be licensed
by some other constituent in the core sentence, e.g. Wenn da was gebucht worden ist,
dann ist das nicht in Ordnung. (cf. 6.1.5).
8.5
Distinction of ON, PRED, ON-MOD, and
PRED-MOD
It is not always trivial to distinguish which constituent is ON, PRED, or ON-MOD for
predicative verbs. For this reason, a few criteria and examples are listed here that can
be of help. Here are some properties of ON and PRED:
1. Typically, PRED occurs in MF, whereas ON occurs in VF of verb-second clauses.
This should be considered for annotation, if no other criterion (as described below)
applies.
125
2. Subject-verb agreement always has to be taken into account. For instance, if the
verb is in plural form, the subject has to be plural as well.
3. If there is a suitable NP that could serve as subject, then this NP is annotated as
subject rather than any other constituent with a different syntactic category (PP,
ADVP, etc.).
For verb-second clauses, it is important to follow these two steps in exactly this order
to stick to the distributional criterion that has been chosen for the PRED/ON distinction:
1. Have a look at the constituent in VF. If it is an NP which might serve as subject
and if it agrees with the verb, annotate it as ON.
2. If it does not agree with the verb, annotate it as PRED (ADJP, ADVP, PP, etc.).
Examples:
1. [ON neue Wortschöpfungen] sind [PRED es] nur.
[PRED es] sind nur [ON neue Wortschöpfungen].
oder sind [PRED es] nur [ON neue Wortschöpfungen].
[PRED das] sind ohnehin [ON die schwächsten Partner].
[ON die schwächsten Partner] sind [PRED das] ohnehin.
oder sind [PRED das] ohnehin [ON die schwächsten Partner].
Subject-verb agreement suggests that neue Wortschöpfungen und die schwächsten
Partner are the subject, because of their plural form regardless in which field they
occur.
2. [ON die Ursache] war [PRED unklar].
[PRED unklar] war [ON die Ursache]
[ON Candan Ercettin] ist [PRED überall].
[PRED überall] ist [ON Candan Ercettin].
ADJPs and ADVPs typically have PRED function when occurring together with
predicative verbs and NP subjects.
3. [PRED aus den Trauernden] wird [ON ein wütender Mop].
ein wütender Mop is considered the subject, because it is a noun phrase. Therefore,
the prepositional phrase is PRED.
4. [ON
[ON
[ON
[ON
das] ist [PRED eine einmalige Chance].
eine einmalige Chance] ist [PRED das].
es] ist [PRED der erste Besuch eines Papstes].
der erste Besuch eines Papstes] ist [PRED es].
126
[ON Hauptauftraggeber] ist [PRED die Bremer Verwaltung].
[ON die Bremer Verwaltung] ist [PRED Hauptauftraggeber].
The NP in VF position agrees with the verb and therefore has subject priority. As
a consequence, the constituent in MF is PRED.
5. [PRED wer] bin [ON ich].
[PRED was] ist [ON das].
In w-questions, the interrogative pronoun is always PRED because here also the
agreement rule applies.
6. [ON-MOD es] sei [PRED wichtig], [ON daß man ... ].
[ON Aufgabe des Festspielhauses] sei [PRED-MOD es], [PRED das Haus spielfertig
zu halten].
If a sentential subject or a sentential predicate occurs with an expletive es, the
expletive es is either ON-MOD or PRED-MOD (cf. 4.2.10).
127
Chapter 9
The TüBa-D/Z Data Formats
The TüBa-D/Z treebank is released in five different data formats :
1. the NEGRA Export format
2. the Penn Treebank format
3. the Export-XML format (incl. anaphora and coreference relations)
4. the TIGER-XML format
5. the CoNLL format
9.1
The NEGRA Export Format
This format is provided by the annotation tool Annotate (Brants and Skut 1998), it is
created automatically from the database underlying the annotation process in Annotate.
The NEGRA Export format is a line-oriented pointer-based representation of the syntactic annotation. It is also the most complete data format since it preserves all the
information available during the manual annotation. A more complete description of the
negra Export format can be found in (Brants 1997).
There are two versions of this format; NEGRA Export format 3 contains 3 layers of token information (token, POS tags, morphology), lemmata are displayed in the ‘comment’
column; NEGRA Export format 4 contains a 4th layer of lemma annotation.
An example of the NEGRA Export format 4 is given below, combined with the graphical representation of the syntactic annotation for the sentence ”Vikare müssen sich nach
dem Kandidatengetz so verhalten, wie es von einem künftigen Pfarrer erwartet werden
kann”.
128
Graphical representation (print out of the annotate tool):
SIMPX
523
−
−
−
−
−
NF
522
PRED−MOD
SIMPX
521
−
−
−
MF
520
ON
FOPP
MF
PX
518
OA
VF
LK
512
PRED
PX
513
ON
519
V−MOD
HD
VC
514
HD
−
NX
515
−
HD
VC
516
OV
−
517
−
HD
OV
OV
refvc
NX
VXFIN
500
NX
501
HD
502
HD
Vikare
−
sich
1
ADVX
503
HD
müssen
0
NX
nach
2
HD
dem
3
504
HD
Kandidatengetz
4
C
505
−
verhalten
6
NX
506
HD
so
5
VXINF
,
507
508
HD
wie
VXINF
509
HD
es
von
einem
11
Pfarrer
13
511
HD
erwartet
14
VXFIN
510
HD
künftigen
12
HD
509
VXINF
HD
werden
15
kann
16
.
8
9
10
NN
VMFIN
PRF
APPR
ART
NN
ADV
VVINF
$,
KOUS
PPER
APPR
ART
ADJA
NN
VVPP
VAINF
VMFIN
$.
npm
3pis
ap*3
d
dsn
dsn
−−
−−
−−
−−
nsn3
d
dsm
dsm
dsm
−−
−−
3sis
−−
LM=Vikar
LM=müssen%aux
LM=#refl
LM=nach
LM=verhalten
LM=,
LM=wie
LM=es
LM=von
LM=ein
LM=künftig
LM=Pfarrer
LM=erwarten
LM=werden%passiv
LM=können%aux
LM=.
LM=dasLM=Kandidatengesetz Kandidatengesetz¦ LM=so
7
ADJX
17
18
The first line of the sentence representation (marked as ’begin of sentence’ (BOS)
includes the sentence id (here: 24538), the identity of the last annotator (here the one
with id 2), the time of the last modification (in UNIX format, i.e. seconds since 1/1/1970)
and the id of the origin of the file (1146 points to article 155 of the edition of 11/7/1992).
In the right column, secondary edges (here: ’refvc’ pointing from node # 518 to node
# 517, a dependency within the verbal complex) as well as corrections of misspellings
(here: ’Kandidatengesetz’) are also represented. Optionally, there is a version of NEGRA
Export format 4 that contains anaphoric relations (here: ’es’ is marked as expletive).
129
Export format 4:
#BOS 24538 2 1134150923 1146
Vikare
Vikar
müssen
müssen%aux
sich
#refl
nach
nach
dem
das
Kandidatengetz
Kandidatengesetz
so
so
verhalten
verhalten
,
,
wie
wie
es
es
von
von
einem
ein
künftigen
künftig
Pfarrer
Pfarrer
erwartet
erwarten
werden
werden%passiv
kann
können%aux
.
.
#500
-#501
-#502
-#503
-#504
-#505
-#506
-#507
-#508
-#509
-#510
-#511
-#512
-#513
-#514
-#515
-#516
-#517
-#518
-#519
-#520
-#521
-#522
-#523
-#EOS 24538
NN
VMFIN
PRF
APPR
ART
NN
ADV
VVINF
$,
KOUS
PPER
APPR
ART
ADJA
NN
VVPP
VAINF
VMFIN
$.
NX
VF
VXFIN
LK
NX
NX
PX
ADVX
MF
VXINF
VC
C
NX
ADJX
NX
PX
MF
VXINF
VXINF
VXFIN
VC
SIMPX
NF
SIMPX
npm
3pis
ap*3
d
dsn
dsn
----nsn3
d
dsm
dsm
dsm
--3sis
--------------------------
HD
500
HD
502
HD
504
506
505
HD
505
HD
507
HD
509
-0
511
HD
512
515
514
HD
513
HD
514
HD
517
HD
518
HD
519
-0
ON
501
523
HD
503
523
OA
508
HD
506
V-MOD
508
PRED
508
523
OV
510
523
521
ON
516
514
HD
515
FOPP
516
521
OV
520
OV
520
HD
520
521
PRED-MOD
523
-0
%% Kandidatengesetz
%% R=expletive
refvc
517
522
The only deviation from context-freeness which the annotation scheme allows concerns the annotation of parentheses. Parentheses are annotated as separate trees with
no attachment to surrounding trees. The following tree gives an example for such a
phenomenon (for a more complete description of the annotation cf. 7.6).
130
Graphical representation:
SIMPX
517
−
−
VF
514
515
−
NX
LK
MF
509
etwas
0
,
1
sagen
2
511
die
3
NX
503
HD
,
5
HD
hätten
6
ADVX
504
HD
Abgeordneten
4
VC
512
−
VXFIN
502
−
MOD
ADVX
HD
NX
501
HD
MOD
LK
ON
VXFIN
500
HD
ON
510
HD
ADVX
So
516
−
HD
−
MF
OA
508
−
−
SIMPX
VXINF
506
507
HD
auch
8
OV
ADVX
505
HD
sie
7
513
HD
HD
noch
9
nicht
10
erlebt
11
.
12
13
ADV
PIS
$,
VVFIN
ART
NN
$,
VAFIN
PPER
ADV
ADV
PTKNEG
VVPP
$.
−−
***
−−
3pis
np*
np*
−−
3pkt
np*3
−−
−−
−−
−−
−−
LM=so
LM=etwas
LM=,
LM=sagen
LM=haben%aux
LM=sie
LM=auch
LM=noch
LM=nicht
LM=erleben
LM=.
LM=der|die|dasLM=Abgeordneter|Abgeordnete|Abgeordnetes¦LM=,
The pointer-based representation of the NEGRA Export format separates information
about the linear precedence of words from attachment information so that parentheses
can be represented naturally without having to resort to explicitly marking non-attached
nodes. Here, the SIMPX node dominating the parenthesis (node #517) is marked as not
having a mother node.
Export format 3:
#BOS 7307 5 1121695339
So
etwas
,
sagen
die
Abgeordneten
,
hätten
sie
auch
noch
nicht
erlebt
.
#500
#501
#502
#503
#504
#505
#506
#507
#508
#509
#510
#511
#512
#513
#514
#515
#516
#517
#EOS 7307
364
ADV
PIS
$,
VVFIN
ART
NN
$,
VAFIN
PPER
ADV
ADV
PTKNEG
VVPP
$.
ADVX
NX
VF
VXFIN
LK
NX
ADVX
ADVX
ADVX
MF
VXINF
VC
SIMPX
VXFIN
LK
NX
MF
SIMPX
-***
-3pis
np*
np*
-3pkt
np*3
------------------------
HD
HD
-HD
HD
-HD
HD
HD
HD
HD
HD
-OA
HD
ON
MOD
MOD
OV
-HD
ON
--
500
501
0
513
515
515
0
503
505
506
507
508
510
0
501
502
512
504
512
509
509
508
509
512
511
512
0
514
517
516
517
0
131
%%
%%
%%
%%
%%
%%
%%
%%
%%
%%
%%
%%
%%
%%
LM=so
LM=etwas
LM=,
LM=sagen
LM=der|die|das
LM=Abgeordneter|Abgeordnete|Abgeordnetes
LM=,
LM=haben%aux
LM=sie
LM=auch
LM=noch
LM=nicht
LM=erleben
LM=.
9.2
The Penn Treebank Format
This format is based on the format of the Penn Treebank (Mitchell et al. 1993). The
attachment of constituents is shown via bracketing and indentation. Thus, all constituents
which show the same level of indentation are attached on the same level. In the Penn
Treebank format, grammatical functions, which are shown in the NEGRA Export format
in the column ”edge label”, are attached to the syntactic label via a colon. Thus, the label
”NX:OA” means that the constituent is a noun phrase with the grammatical function
accusative object.
The Penn Treebank format is a representation that combines the linear representation
of words with their attachment to higher constituents. For this reason, this format is
restricted to completely context-free tree structures, i.e. it cannot adequately represent
the annotation of parentheses in TüBa-D/Z. In order to capture the original syntactic
annotation as well as the original word order in the sentence, it was decided to introduce
a new edge label to mark such cases: PAR. Thus, the sentence ”So etwas , sagen die
Abgeordneten , hätten sie auch noch nicht erlebt .”, as shown above is represented in the
Penn Treebank format by the following bracketed structure:
Comments are preceded by a double ’%’ sign. The comment behind the structure is
intended to help the reader locate the beginning of the parenthesis and it is not part of
the actual data.
132
%% sent. no. 7307
(
(SIMPX
(VF
(NX:OA
(ADVX
(ADV:HD So)
)
(PIS:HD etwas)
)
)
($, ,)
(SIMPX:PAR
(LK
(VXFIN:HD
(VVFIN:HD sagen)
)
)
(MF
(NX:ON
(ART die)
(NN:HD Abgeordneten)
)
)
)
($, ,)
(LK
(VXFIN:HD
(VAFIN:HD hätten)
)
)
(MF
(NX:ON
(PPER:HD sie)
)
(ADVX:MOD
(ADV:HD auch)
)
(ADVX:MOD
(ADVX
(ADV:HD noch)
)
(PTKNEG:HD nicht)
)
)
(VC
(VXINF:OV
(VVPP:HD erlebt)
)
)
)
($. .)
)
%% here starts the parenthesis!
133
Commas, which are not attached to the tree, are indented on the highest level although
they are included in the bracketing of the constituent surrounding them. In the sentence
below, e.g., the first comma is grouped into the noun phrase NX via word order. The
indentation, however, signals that the comma cannot necessarily be attached to this node.
It is also conceivable that it may be attached to one of the lower nodes, NX or R-SIMPX.
In the case of the second comma, there are even more possible attachment sites.
%% fragment of sent. no. 33
(
(R-SIMPX
(C
(NX:ON
(PRELS:HD die)
)
)
(MF
(NX:OA
(NX=ORG:HD
(ART:-NE die)
(NN:HD AWO)
)
($, ,)
(R-SIMPX
(C
(PX:V-MOD
(PWAV:HD wo)
)
)
(MF
(NX:ON
(PPER:HD er)
)
(NX:PRED
(NN:HD Kreisvorsitzender)
)
)
(VC
(VXFIN:HD
(VAFIN:HD ist)
)
)
)
)
)
($, ,)
(VC
(VXFIN:HD
(VVFIN:HD prüfte)
)
)
)
($. .))
134
9.3
The Export-XML Format
The XML format is a custom-made XML format that follows the NEGRA Export file
format. It is designed to accommodate all original information provided in the Export
format, including e.g. comments . Dominance relations between nodes are represented
directly within the XML tree structure. Nodes without parent node (e.g. the sentence
node SIMPX and punctuation marks do not have any ”parent” attribute. They have
the edge label ”- -” that can be linked to an implicit root node. Thus, it is possible to
represent parentheses without the use of additional labels.
Anaphora is expressed by a link between two related nodes. Coreference sets therefore
are represented implicitly by chains of nodes that are part of a referential relation.
The following example shows the XML structure for the sentence “Schillen erklärte,
sie werde als Kriegsgegnerin kandidieren”. The personal pronoun “sie” is anaphoric to the
antecedent noun phrase “Schillen”. In the XML document, a <relation> tag is added
below each node that is part of a referential relation. It encodes the type of referential
relation and the node ID of the antecedent node. In our example, the antecedent is the
node with ID s1723 500, that is the NX dominating the named entity “Schillen”. This
NX in turn is in a coreferential relationship with node s1721 11 (word number 11 in
sentence 1721), thus part of a coreference chain.
Please note that the format of Export-XML has changed from release 6 to release 7
of the TüBa-D/Z. There are XMLPerl skripts available for the old Export-XML format
and a Java library of the new Export-XML format.
Graphical representation of the tree without annotation of the referential relation:
SIMPX
514
−
−
−
NF
513
OS
SIMPX
512
−
VF
LK
506
VF
507
ON
HD
NX
VXFIN
500
Schillen
NX
VXFIN
HD
erklärte
0
NE
HD
,
1
VVFIN
2
$,
sie
511
OV
NX
503
VXINF
504
−
werde
3
VC
510
PRED
HD
PPER
−
MF
509
ON
502
HD
−
LK
508
501
HD
−
als
4
VAFIN
505
HD
HD
Kriegsgegnerin
5
KOKOM
kandidieren
6
NN
.
7
VVINF
8
$.
nsf
3sit
−−
nsf3
3sks
−−
nsf
−−
−−
LM=Schillen
LM=erklären
LM=,
LM=sie
LM=werden%aux
LM=als
LM=Kriegsgegnerin
LM=kandidieren
LM=.
135
XML format including the referential relation:
<sentence xml:id="s1723">
<node xml:id="s1723_514" cat="SIMPX" func="--">
<node xml:id="s1723_501" cat="VF" func="-" parent="s1723_514">
<node xml:id="s1723_500" cat="NX" func="ON" parent="s1723_501">
<relation type="coreferential" target="s1721_11"/>
<ne xml:id="ne_29507" type="PER">
<word xml:id="s1723_1" form="Schillen" pos="NE" morph="nsf" lemma="Schillen" func="HD" parent="s1723_500"/>
</ne>
</node>
</node>
<node xml:id="s1723_503" cat="LK" func="-" parent="s1723_514">
<node xml:id="s1723_502" cat="VXFIN" func="HD" parent="s1723_503">
<word xml:id="s1723_2" form="erklärte" pos="VVFIN" morph="3sit" lemma="erklären" func="HD" parent="s1723_502"/>
</node>
</node>
<word xml:id="s1723_3" form="," pos="$," lemma="," func="--"/>
<node xml:id="s1723_513" cat="NF" func="-" parent="s1723_514">
<node xml:id="s1723_512" cat="SIMPX" func="OS" parent="s1723_513">
<node xml:id="s1723_505" cat="VF" func="-" parent="s1723_512">
<node xml:id="s1723_504" cat="NX" func="ON" parent="s1723_505">
<relation type="anaphoric" target="s1723_500"/>
<word xml:id="s1723_4" form="sie" pos="PPER" morph="nsf3" lemma="sie" func="HD" parent="s1723_504"/>
</node>
</node>
<node xml:id="s1723_507" cat="LK" func="-" parent="s1723_512">
<node xml:id="s1723_506" cat="VXFIN" func="HD" parent="s1723_507">
<word xml:id="s1723_5" form="werde" pos="VAFIN" morph="3sks" lemma="werden%aux" func="HD" parent="s1723_506"/>
</node>
</node>
<node xml:id="s1723_509" cat="MF" func="-" parent="s1723_512">
<node xml:id="s1723_508" cat="NX" func="PRED" parent="s1723_509">
<word xml:id="s1723_6" form="als" pos="KOKOM" lemma="als" func="-" parent="s1723_508"/>
<word xml:id="s1723_7" form="Kriegsgegnerin" pos="NN" morph="nsf" lemma="Kriegsgegnerin" func="HD"
parent="s1723_508"/>
</node>
</node>
<node xml:id="s1723_511" cat="VC" func="-" parent="s1723_512">
<node xml:id="s1723_510" cat="VXINF" func="OV" parent="s1723_511">
<word xml:id="s1723_8" form="kandidieren" pos="VVINF" lemma="kandidieren" func="HD" parent="s1723_510"/>
</node>
</node>
</node>
</node>
</node>
<word xml:id="s1723_9" form="." pos="$." lemma="." func="--"/>
</sentence>
136
9.4
The TIGER-XML Format
The TIGER-XML format was designed by Lezius (2002) as interchange format for the
TIGERSeach treebank query tool. The corpus body contains sentences ”s” containing
one or several “graphs”. Graphs are annotated stand-off: A list of terminal nodes is
followed by nonterminals with their edges.
TIGER-XML doesn’t include the full set of annotations. POS, morphology, and
lemmas are included; corrected word forms and anaphora are missing.
A detailed description is provided in chapter V “The TIGER-XML treebank encoding
format” of the TIGERSearch User’s Manual (König et al. 2003).
137
Sentence 1723 in TIGER-XML format:
<s id="s1723">
<graph root="s1723_514">
<terminals>
<t id="s1723_1" word="Schillen" lemma="Schillen" pos="NE" morph="nsf" />
<t id="s1723_2" word="erklärte" lemma="erklären" pos="VVFIN" morph="3sit" />
<t id="s1723_3" word="," lemma="," pos="$," morph="--" />
<t id="s1723_4" word="sie" lemma="sie" pos="PPER" morph="nsf3" />
<t id="s1723_5" word="werde" lemma="werden%aux" pos="VAFIN" morph="3sks" />
<t id="s1723_6" word="als" lemma="als" pos="KOKOM" morph="--" />
<t id="s1723_7" word="Kriegsgegnerin" lemma="Kriegsgegnerin" pos="NN" morph="nsf" />
<t id="s1723_8" word="kandidieren" lemma="kandidieren" pos="VVINF" morph="--" />
<t id="s1723_9" word="." lemma="." pos="$." morph="--" />
</terminals>
<nonterminals>
<nt id="s1723_500" cat="NX=PER">
<edge label="HD" idref="s1723_1" />
</nt>
<nt id="s1723_501" cat="VF">
<edge label="ON" idref="s1723_500" />
</nt>
<nt id="s1723_502" cat="VXFIN">
<edge label="HD" idref="s1723_2" />
</nt>
<nt id="s1723_503" cat="LK">
<edge label="HD" idref="s1723_502" />
</nt>
<nt id="s1723_504" cat="NX">
<edge label="HD" idref="s1723_4" />
</nt>
<nt id="s1723_505" cat="VF">
<edge label="ON" idref="s1723_504" />
</nt>
<nt id="s1723_506" cat="VXFIN">
<edge label="HD" idref="s1723_5" />
</nt>
<nt id="s1723_507" cat="LK">
<edge label="HD" idref="s1723_506" />
</nt>
<nt id="s1723_508" cat="NX">
<edge label="-" idref="s1723_6" />
<edge label="HD" idref="s1723_7" />
</nt>
<nt id="s1723_509" cat="MF">
<edge label="PRED" idref="s1723_508" />
</nt>
<nt id="s1723_510" cat="VXINF">
<edge label="HD" idref="s1723_8" />
</nt>
<nt id="s1723_511" cat="VC">
<edge label="OV" idref="s1723_510" />
</nt>
<nt id="s1723_512" cat="SIMPX">
<edge label="-" idref="s1723_505" />
<edge label="-" idref="s1723_507" />
<edge label="-" idref="s1723_509" />
<edge label="-" idref="s1723_511" />
</nt>
<nt id="s1723_513" cat="NF">
<edge label="OS" idref="s1723_512" />
</nt>
<nt id="s1723_514" cat="SIMPX">
<edge label="-" idref="s1723_501" />
<edge label="-" idref="s1723_503" />
<edge label="-" idref="s1723_513" />
</nt>
</nonterminals>
</graph>
</s>
138
9.5
The CoNLL Format
CoNLL format contains a dependency version of TüBa-D/Z in the format of the CoNLLX shared task. The conversion was done automatically, but is oriented at the annotation
guidelines by Foth (2006).
CoNLL format is a table format containing a series of tabular-separated lines. Each
line contains the following information:
1. ID – a sequential ID for each token
2. FORM – the word form (token)
3. LEMMA – the gold standard lemma of the token
4. CPOSTAG – simplified part-of-speech tag
5. POSTAG – part-of-speech tag according to STTS tag set
6. FEATS – tag with morphological information
7. HEAD – regent of the token in the dependency analysis, or “0” for tokens without
regent
8. DEPREL – dependency relation between the token and its regent, or “ROOT” for
tokens without regent
Sentence 1723 in CoNLL format:
1
2
3
4
5
6
7
8
9
Schillen
erklärte
,
sie
werde
als
Kriegsgegnerin
kandidieren
.
Schillen
erklären
,
sie
werden%aux
als
Kriegsgegnerin
kandidieren
.
N
V
$,
PRO
V
KOKOM
N
V
$.
139
NE
VVFIN
$,
PPER
VAFIN
KOKOM
NN
VVINF
$.
nsf
3sit
-nsf3
3sks
-nsf
---
2
0
2
5
2
8
6
5
8
SUBJ
ROOT
-PUNCTSUBJ
S
KOM
CJ
AUX
-PUNCT-
References
Bech, G. 1955–57. Studien über das deutsche Verbum infinitum. Kopenhagen. 2 Bände.
2. unveränderte Auflage 1983 mit einem Vorwort von Catharine Fabricius-Hansen.
Tübingen: Max Niemeyer.
Behaghel, O. 1932. Deutsche Syntax (Eine geschichtliche Darstellung), Band 4. Heidelberg: Carl Winter.
Brants, T., and W. Skut. 1998. Automation of treebank annotation. In Proceedings
of the Conference on New Methods in Language Processing (NeMLaP-3/CoNLL98),
January 14-17, 1998, Sydney, Australia, 49–57. Sydney.
Brants, T. 1997. The NeGra Export Format for Annotated Corpora. Universität des
Saarlandes, Computational Linguistics, Saarbrücken, Germany.
Brants, T. 1998. TnT – A Statistical Part-of-Speech Tagger. Universität des Saarlandes,
Computational Linguistics, Saarbrücken, Germany.
Bresnan, J. 1995. Lexicality and Argument Structure. In Invited Paper given at the
Paris Syntax and Semantics Conference. Paris. October 12-14, 1995. URL: http:
//www-csli.stanford.edu/bresnan/download.html.
Drach, E. 1937. Grundgedanken der Deutschen Satzlehre. Frankfurt/Main.
Drosdowski, G. (Ed.). 1995. Duden ”Die Grammatik der deutschen Gegenwartssprache”.
Mannheim, Leipzig, Wien, Zürich: Dudenverlag.
Eisenberg, P. 1999–2001. Grundriß der deutschen Grammatik, Band 2: Der Satz.
Stuttgart, Weimar: J.B. Metzler.
Engel, U. 1996. Deutsche Grammatik. Heidelberg: Julius Groos Verlag.
Erdmann, O. 1886. Grundzüge der deutschen Syntax nach ihrer geschichtlichen Entwicklung dargestellt. Stuttgart: Cotta. Erste Abteilung.
Foth, K. A. 2006. Eine umfassende Constraint-Dependenz-Grammatik des Deutschen.
Technical report, Fachbereich Informatik der Universität Hamburg.
Grewendorf, G. 1991. Aspekte der deutschen Syntax. Vol. 33 of Studien zur deutschen
Grammatik. Tübingen: Gunter Narr Verlag.
Helbig, G., and J. Buscha. 1998. Deutsche Grammatik. Ein Handbuch für den
Ausländerunterricht. Leipzig: Verl. Enzyklopädie. 18 edition.
Herling, S. H. A. 1821. Über die Topik der deutschen Sprache. In Abhandlungen des frankfurterischen Gelehrtenvereins für deutsche Sprache, 296–362, 394. Frankfurt/Main.
Drittes Stück.
140
Hinrichs, E. W., and H. Telljohann. 2009. Constructing a Valence Lexicon for a Treebank
of German. In Proceedings of the Seventh International Workshop on Treebanks and
Linguistic Theories (TLT 7): January 23-24, 2009, Groningen, The Netherlands.
URL: http://www.let.rug.nl/tlt/.
Hinrichs, E. W., J. Bartels, Y. Kawata, V. Kordoni, and H. Telljohann. 2000. The
Tübingen Treebanks for Spoken German, English, and Japanese. In W. Wahlster
(Ed.), Verbmobil: Foundations of Speech-to-Speech Translation. Berlin: Springer.
Höhle, T. N. 1986. Der Begriff ‘Mittelfeld’. Anmerkungen über die Theorie der topologischen Felder. In A. Schöne (Ed.), Kontroversen alte und neue. Akten des 7. Internationalen Germanistenkongresses Göttingen, 329–340. Tübingen: Niemeyer.
Kathol, A. 1995. Linearization-Based German Syntax. PhD thesis, Ohio State University.
Kiss, T. 1995. Infinitive Komplementation. Neue Studien zum deutschen Verbum infinitum. Tübingen: Max Niemeyer.
König, E., W. Lezius, and H. Voormann. 2003. TIGERSearch 2.1 – User’s Manual. URL: http://www.ims.uni-stuttgart.de/projekte/TIGER/TIGERSearch/
manual.shtml.
Kübler, S., and H. Telljohann. 2002. Towards a dependency-based evaluation for partial
parsing. In Beyond PARSEVAL – Towards Improved Evaluation Measures for Parsing
Systems – (LREC 2002 Workshop), Las Palmas, Gran Canaria, June 2002.
Lezius, W. 2002. Ein Suchwerkzeug für syntaktisch annotierte Textkorpora. PhD thesis,
IMS, University of Stuttgart. URL: http://www.ims.uni-stuttgart.de/projekte/
TIGER/TIGERSearch/manual_html.html.
Mitchell, M., B. Santorini, and M. A. Marcinkiewicz. 1993. Building a Large Annotated
Corpus of English: The Penn Treebank. Computational Linguistics 19(2):313–330.
Naumann, K., and V. Möller. 2007. Manual for the Annotation of in-document Referential
Relations. University of Tübingen, May 2007.
Plaehn, O. 1998. Annotate – Bedienungsanleitung. FR 8.7 Computerlinguistik, Projekt C3 Nebenläufige Grammatische Verarbeitung, Sonderforschungsbereich 378,
Ressourcenadaptive Kognitive Prozesse, 13. April 1998. URL: http://www.coli.
uni-saarland.de/projects/sfb378/negra-corpus/annotate.html.
Pütz, H. 1986. Über die Syntax der Pronominalform ’es’ im modernen Deutsch. Tübingen:
Stauffenburg. 2 edition.
Schiller, A., S. Teufel, and C. Thielen. 1995. Guidelines für das Tagging deutscher Textcorpora mit STTS. Technical report, Universities of Stuttgart and Tübingen. URL:
http://www.sfs.nphil.uni-tuebingen.de/ELWIS/stts/stts.html.
Stegmann, R., H. Telljohann, and E. W. Hinrichs. 2000. Stylebook for the German
Treebank in verbmobil. Verbmobil-report 239, University of Tübingen.
141
Telljohann, H., E. W. Hinrichs, S. Kübler, H. Zinsmeister, and K. Beck. 2009. Stylebook
for the Tübingen Treebank of Written German (TüBa-D/Z). University of Tübingen,
November 2009.
Trushkina, J. 2004. Morpho-syntactic annotation and dependency parsing of German.
PhD thesis, University of Tübingen. URL: http://w210.ub.uni-tuebingen.de/
dbt/volltexte/2004/1523).
Versley, Y., K. Beck, E. W. Hinrichs, and H. Telljohann. 2010. A Syntax-first Approach
to High-quality Morphological Analysis and Lemma Disambiguation for the TüBaD/Z Treebank. In Proceedings of the Ninth International Workshop on Treebanks
and Linguistic Theories (TLT 9): December 3-4, 2010, Tartu, Estonia. URL: http:
//www.math.ut.ee/tlt9/.
142
Index
accusative object, double, 73
AcI, 72
adverbial adjective, 13, 58, 60
adverbial phrase, 17, 63
ambiguity, 10, 23, 25, 80, 82–85, 103
apposition, 18, 34
attributive adjective, 13, 24, 27, 28, 58, 61
isolated phrase, 20, 21, 84, 85, 107
C-field, 8, 9, 17, 86, 89, 92
cardinal numbers, 13, 32, 48, 49
circumposition, 13, 58
coherency, 71, 72
comparatives, 5, 112
CoNLL format, 128
context-freeness, 4
coordination, 5, 7, 12, 17, 29, 56, 61, 95–99,
101–108, 114
modal verbs, 14, 78
KOORD-field, 8, 9, 17, 88, 95
lassen, 72
levels of annotation, 11
long-distance dependency, 4, 23, 90, 113
longest match principle, 10, 14, 22, 105
Dependency Grammar, 24, 55
determiner phrase, 17, 54
discourse marker, 3, 5, 14, 17, 24, 119, 120
named entities, 4, 12, 19, 33, 41–44, 47
Negra export format, 128
Negra treebank, 2
node labels, 4, 11, 12, 17, 19, 38, 67, 85,
102, 119
nominalized adjective, 60
non-ambiguity, 23, 81
non-words, 14, 50
ordinal numbers, 48
paratactic construction, 17, 105, 107
parenthesis, 3, 5, 14, 121
PARORD-field, 8, 9, 17, 88, 89, 107
part-of-speech tags, 4, 11, 20, 50, 56
particle verb, 75
Penn Treebank format, 128
postmodifier, 10, 25, 31, 33, 35, 40, 44, 46,
63, 83, 93, 94, 112
postnominal modifier, 31, 32
flat clustering principle, 10, 25, 65, 86
postposition, 13, 58
foreign language material, 13, 38, 47
predicate, 75, 76, 123, 127
predicate-argument structure, 3, 9
headline, 3, 5, 14, 67, 80, 81, 117
high attachment principle, 10, 25, 101, 114 predicative adjective, 13, 58, 59, 115
preference principle, 81, 117
premodifier, 25, 28, 44–46, 48, 59, 60, 62,
imperative, 14, 73
63, 83, 103
incoherency, 72
prenominal modifier, 27, 29, 31
infinitives with zu, 69, 70, 72
preposition, 13, 24, 26, 55, 56, 124
initial field, 6, 8, 17, 83
edge labels, 4, 9, 11, 12, 18–20, 23, 24, 81,
95
elliptical construction, 5, 11, 14, 95, 109,
111, 118
Ersatzinfinitiv, 8, 17, 67, 68
expletive, 18, 51, 52, 127
export-XML format, 128
143
proper noun, 13, 19, 25, 26, 29, 37, 38, 41,
47
punctuation marks, 14, 16, 34, 83, 84
relative clause, 8, 17, 82, 92–94
relative clause, event-modifying, 94
relative clause, independent, 94
resumptive construction, 7, 8, 17, 90, 125
reusability, 2, 3
secondary edge label, 4, 12, 18, 23, 65, 66,
72, 93
split coordination, 18, 96, 108
superlative forms, 112
syntactic dependencies, 4
syntactic-semantic node labels, 12, 42, 43
TüBa-D/S treebank, 2
TüBa-D/Z data formats, 2, 128
theory-neutrality, 2, 3
TIGER treebank, 2
TigerXML format, 128
topicalization, 5, 116
topological fields, 4–9, 11, 12, 23, 24, 27, 76,
80, 86, 102, 109, 111
truncated word, 14, 99
verb complex, 4, 6–8, 17, 65–67, 69, 71, 72,
75, 93, 116
verb particle, 14, 74
VERBMOBIL treebank, 1, 2, 20
144