Stylebook for the T¨ubingen Treebank of Written German (T¨uBa-D/Z) Heike Telljohann, Erhard W. Hinrichs, Sandra K¨ ubler, Heike Zinsmeister, Kathrin Beck Universit¨at T¨ ubingen Seminar f¨ ur Sprachwissenschaft Wilhelmstr. 19 D-72074 T¨ ubingen {telljohann,eh,kuebler,zinsmeis,kbeck}@sfs.uni-tuebingen.de November 2009
Abstract This stylebook is an updated version of Telljohann et al. (2006). It describes the design principles and the annotation scheme for the German treebank T¨ uBa-D/Z developed by the Division of Computational Linguistics (Lehrstuhl Prof. Hinrichs) at the Department of Linguistics (Seminar f¨ ur Sprachwissenschaft – SfS) of the Eberhard Karls Universit¨at T¨ ubingen, 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 45,200 sentences. The newspaper material is taken from the taz editions from 1992 July 10, 11, 13, 14 1995 October 14, 16, 17 1999 April 30, May 3 – 7 The average sentence length is 17.6 words and the total number of tokens currently amounts to 794,079. The T¨ uBa-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/de_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 T¨ uBa-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 T¨ uBa-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 theory-neutrality.
0
Acknowledgements The annotation of the T¨ uBa-D/Z treebank is carried out as part of the Competence Center for Text- and Information Technology (or officially: Kompetenzzentrum f¨ ur Text- und Informationstechnologie – KIT). KIT is a joint project of the Institute for Natural Language Processing (IMS) in Stuttgart, and the ’Seminar f¨ ur Sprachwissenschaft’ (SfS) in T¨ ubingen. Funding has been provided since 2000 by the Ministry of Science, Research and the Arts Baden-W¨ urttemberg. Additional support for the creation of the T¨ uBa-D/Z treebank was provided by the special research center Linguistic Data Structures (Sonderforschungsbereich Linguistische Datenstrukturen – SFB 441) at the University of T¨ ubingen funded by the German Research Council (Deutsche Forschungsgemeinschaft – DFG). 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 (see Trushkina 2004) and Yannick Versley who provided the tools for morphological preprocessing. 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. M¨ uller in the course of their research work in the SFB 441. They deserve special mention for their support. We like to thank Vera M¨oller and Karin Naumann (see (Naumann and M¨oller 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 course of their PhD projects in the SFB 441 they enhanced the conceptual aspects of the anaphora resolution as annotated in the treebank. They also 1
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, Silke Dutz, Katrin Eichler, Emilia Ellsiepen, Steffen Froemel, Holger Gauza, Simone Hartung, Daniel H¨ uttl, Heike Johannsen, Miriam K¨ashammer, Laura Kassner, Sarah Klug, Janina Kopp, Christian Kreß, Rebecca Kreß, Michael Kossack, Anne Lohse, Wolfgang Maier, Nicole Maruschka, Kai Metzger, Vera M¨oller, Simone M¨ uller, Maja Pietsch, Andreas Rudin, Maria Schmidt, Marie Schreier, Insa Starr, Melanie St¨orzer, and Dominikus Wetzel. They also improved the linguistic quality of the annotation by dedicated discussions on problematic and interesting examples.
2
The development of the T¨ uBa-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 ’Universit¨at des Saarlandes’ 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) Nebenl¨ aufige grammatische Verarbeitung (NEGRA) in the 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 T¨ ubingen treebank project.
3
Contents List of Tables
7
1 Introduction
8
2 Major Challenges and Design Decisions
10
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 . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
13 13 13 16 17 17 17 17 17 17 18 23 26 27 29 30
4 The Annotation of the Internal Structure of Phrases 4.1 Premodification and Postmodification in Phrases . . . 4.2 Noun Phrases . . . . . . . . . . . . . . . . . . . . . . . 4.2.1 Prenominal Modification . . . . . . . . . . . . 4.2.2 Postnominal Modification . . . . . . . . . . . . 4.2.3 Appositional Constructions . . . . . . . . . . . 4.2.4 Foreign Language Material . . . . . . . . . . . 4.2.5 Proper Nouns and Named Entities . . . . . . . 4.2.6 Ordinal Numbers . . . . . . . . . . . . . . . . . 4.2.7 Cardinal Numbers . . . . . . . . . . . . . . . . 4.2.8 Letters and Non-Words . . . . . . . . . . . . . 4.2.9 Expletive and Other Uses of es . . . . . . . . . 4.3 Determiner Phrases . . . . . . . . . . . . . . . . . . . 4.4 Prepositional Phrases . . . . . . . . . . . . . . . . . . 4.4.1 Prepositions . . . . . . . . . . . . . . . . . . . 4.4.2 Circumpositions and Postpositions . . . . . . . 4
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
31 31 31 31 35 38 42 45 54 55 56 57 60 61 61 64
. . . . . . . . . . . . . . .
4.5 4.6 4.7
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 . . . . . . 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 . . . . . . . . . . . . . . . . . . . 6.5 Elliptical Constructions . . . . . . . . . . . . . . . . . . . . . 5
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
65 70 71 71 71 73 75 77 78 79 81 82 85
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
86 86 86 87 89 89 90
. . . . . . . . . . . . . . . . . . . . . . . . .
92 92 92 94 94 95 96 97 97 98 99 100 100 101 102 103 104 106 107 108 109 111 113 113 114 116
. . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . .
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 . . . . . . . . . . . . . . . . . . . . . . 7.7 Elliptical weil and wenn auch Constructions . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
8 Criteria for the Distinction of Grammatical Functions 8.1 Subcategorization of Verbs . . . . . . . . . . . . . . . . . . . . . . . 8.1.1 Distinction of FOPP, OPP, and V-MOD . . . . . . . . . . . . 8.1.2 Distinction of MOD, MOD-MOD, and V-MOD . . . . . . . . 8.1.3 Distinction of ON, PRED, ON-MOD, and PRED-MOD . . .
. . . . . . . . . . . . .
. . . . . . . . .
119 119 119 119 122 123 124 125 128 129
. . . .
131 131 132 132 133
References
135
Appendix: The T¨ uBa-D/Z Data Formats
137
Index
144
6
List of Tables 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8
Three clause types according to H¨ohle (1986) . . . . . Topological fields . . . . . . . . . . . . . . . . . . . . Levels of annotation . . . . . . . . . . . . . . . . . . Morphological feature combinations for lexical tokens Values of morphological features . . . . . . . . . . . . Node labels . . . . . . . . . . . . . . . . . . . . . . . Edge labels . . . . . . . . . . . . . . . . . . . . . . . Labels for proper nouns and named entities . . . . .
. . . . . . . .
14 15 18 22 23 24 25 26
4.1
Types of es . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
59
7
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
Chapter 1 Introduction The purpose of this report is to describe the design principles and annotation scheme for the T¨ uBa-D/Z treebank of German. It is intended as a guide for the treebank annotators in T¨ ubingen 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 T¨ uBa-D/Z treebank was developed by the Division of Computational Linguistics (Lehrstuhl Prof. Hinrichs) at the Department of Linguistics (Seminar f¨ ur Sprachwissenschaft – SfS) of the Eberhard Karls Universit¨at T¨ ubingen, 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 45,200 sentences. The newspaper material is taken from the taz editions from 1992 July 10, 11, 13, 14 1995 October 14, 16, 17 1999 April 30, May 3 – 7. The average sentence length is 17.6 words and the total number of tokens currently amounts to 794,079. The T¨ uBa-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/de_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 T¨ uBa-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 annotation scheme for the verbmobil treebank has been summarized in the ’Stylebook for the German Treebank in verbmobil’ (Stegmann et al. 2000). The T¨ uBa-D/Z annotation scheme has been extended along various dimensions to accommodate the characteristics of written texts. 8
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 T¨ uBa-D/Z treebank is released in three different data formats : the Negra export format, the Penn treebank format, and in XML format. More information about each data format is given in Appendix: The TBa-D/Z Data Formats. To the best of our knowledge, the verbmobil treebank for spoken German is still the only treebank based on German speech data. It is released as T¨ uBa-D/S treebank (http://www.sfs.uni-tuebingen.de/en/de_tuebads.shtml). For written texts, T¨ uBa-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 Tbingen Partially Parsed Corpus of Written German (TPP-D/Z; http://www. sfs.uni-tuebingen.de/en/de_tuepp.shtml) is a project closely related to the TBaD/Z treebank. It consists of 200 million word tokens of the Science CD (WissenschaftsCD) of ’die tageszeitung’ (taz), including the sentences which are annotated in the TBaD/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 TBa-D/Z (12/2003) functioned as training corpus.
9
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 no 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 T¨ uBa-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. To support the use of the T¨ uBa-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 10
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 H¨ohle (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 41 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 StuttgartT¨ ubingen tagset, 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. 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 11
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. The stylebook concludes with a bibliography, a subject index, and an appendix which describes the three different data formats in which the T¨ uBa-D/Z treebank is distributed. We do not consider the annotation level of anaphora and coreference relations in this stylebook. Please consult (Naumann and M¨oller 2007) for a detailed description of these phenomena.
12
Chapter 3 The Theoretical Basis of the Annotation Scheme 3.1
Topological Fields
The annotation scheme for the T¨ uBa-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, H¨ohle (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 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 13
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. H¨ohle (1986) denotes the three clause types as E-S¨atze (verb-final clauses), F1S¨atze (verb-initial clauses), and F2-S¨atze (verb-second clauses). The topological schemes of these types are listed in Table 3.1. Table 3.1: Three clause types according to H¨ohle (1986) E-S¨atze (KOORD) - (C) - X - VK - Y F1-S¨atze (KOORD - (KL) - FINIT - X - VK - Y F2-S¨atze (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 H¨ohle (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 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. 1 2
In the following, the abbreviations for the fields listed in Table 3.2 are used. In the fifth release, 52.5% of all Vorfeld fields host the subject.
14
Table 3.2: Topological fields Field VF LK MF VC NF LV C KOORD
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) PARORD Koordinationsfeld (field for non-coordinating particles) left-most element, optionally only in verb-second (e.g. denn, weil)
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 non-verbal 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 u ¨berlegen m¨ ussen) (cf. 4.7.3). NF: For some clause types (e.g. so daß-S¨atze), 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 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 w¨are 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, 15
obwohl, wenn and also by complex interrogative or relative phrases, e.g. ..., ’um wieviel Geld’ geht es dabei? / ..., ’an der’ Max Daniel Professor f¨ ur 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 verb-final clauses is described in H¨ohle (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 catalog 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. 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. 16
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.
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 H¨ohle (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 : 3
We do not consider the suprasentential annotation level of anaphora and coreference relations in this stylebook. Please consult (Naumann and M¨oller 2007) for a detailed description of these phenomena.
17
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 and edge labels for grammatical functions lexical entries tagged with the part-of-speech (PoS-)tags taken from the STTS tagset (Schiller et al. 1995) and with morphological features (Trushkina 2004)
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-T¨ ubingen 4 tagset (STTS) (Schiller et al. 1995). 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 tagset consists of the tags listed in Table 3.4.2 (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). 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 3.4 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 3.5 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 3.6 lists all node labels which are used in the treebank. (An additional node is introduced for named entities, see Table 3.8) 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 3.7, including secondary edge labels. The latter ones are used to resolve ambiguities on a different level of description. 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.
18
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. (K¨ ubler and Telljohann 2002)). In order to mark proper nouns, and named entities within the treebank, the node label (EN-ADD) and the secondary edge label (EN) are defined (see Table 3.8). EN-ADD is inserted between two nodes to indicate that the node below represents a complex proper noun, (e.g. Ute Wedemeier, The Jim Wane Swingtett), a single proper noun tagged as NN with respect to the STTS (e.g. S¨ ogestraße), or a named entity (e.g. Auf die st¨ urmische Art, Built to Spill) (cf. 4.2.5). EN-ADD is either directly attached to a head noun of a phrase or to a field. If it has a postmodifier, its mother node is NX which represents the nominal status of EN-ADD. The internal syntactic structure of EN-ADD is governed by the general annotation rules. The secondary edge label EN gives information about the relation of two parts of a proper noun within a complex phrase consisting, for instance, of an article and/or an attributive adjective which do not belong to the proper noun itself, e.g. der [zweite Weltkrieg EN] (cf. 4.2.5).
19
Table 3.3.1: The STTS tagset 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 relative pronoun substituting attributive reflexive personal pronoun substituting interrogative pronoun attributive interrogative pronoun
KOUS KON KOKOM NN NE PDS PDAT PIS PIAT PIDAT PPER PPOSS PPOSAT PRELS PRELAT PRF PWS PWAT
20
examples [das] große [Haus] [er f¨ahrt] 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 [M¨anner], [im Jahre] 1994 [Er hat das mit “] A big fish [” u ¨bersetzt] 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 [Br¨ uder] 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]
POS = PWAV PROP PTKZU PTKNEG PTKVZ PTKANT PTKA TRUNC VVFIN VVIMP VVINF VVIZU VVPP VAFIN VAIMP VAINF VAPP VMFIN VMINF VMPP XY $, $. $(
description adverbial interrogative or relative pronoun pronominal adverb 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
21
examples warum, wo, wann, wor¨ uber, wobei daf¨ ur, dabei, deswegen, trotzdem zu [gehen] nicht [er kommt] an, [er f¨ ahrt] rad ja, nein, danke, bitte am [sch¨onsten], 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 d¨ urfen wollen [er hat] gekonnt D2XW3, letters , . ? ! ;: -[]()
Table 3.4: Morphological feature combinations for lexical tokens POS ADJA
feature combination case number gender
APPR
case
comments underspecified for gender if plural noun is underspecified, e.g. die/np* nordhessischen/np* Gr¨ unen/np* invariant local description e.g. Berliner/*** cardinal numbers as abbreviation: full morphology e.g. im 4./dsn Jahrhundert/dsn without case if a prepositions takes another PP as complement, e.g. bis/ zu/d einer/dsf Woche/dsf
APPRART case number, gender APPO ART NE NN
case case number gender case number gender case number gender
PDAT PDS PIAT
case number gender case number gender case number gender
PIDAT
case number gender
PIS
case number gender
PPER
case number gender person case number gender case number gender case number gender case number gender plural is underspecified for gender case number gender sich: underspecified for gender person case number gender plural is underspecified for gender wessen/*** case number gender underspecified for gender: plural forms and wer, wem, wen
PPOSAT PPOSS PRELAT PRELS PRF PWAT PWS
underspecified for gender, e.g. Abgeordnete (in plural), Leute
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 underspecified: man/ns* nichts/*** (cf. nix, sowas) PIS or PIAT: allerhand/*** (cf. allerlei, allzuviel, dergleichen, derlei, etwas, genausoviel, genug, gengend, keinerlei, mehr, reichlich, soviel, viel, wenig, weniger, zuviel, zuwenig) PIDAT or PIS: sowas/*** (cf. paar, bißchen)
22
POS VAFIN VAIMP VMFIN VVFIN VVIMP
feature combination comments person number mood tense person number person number mood tense person number mood tense number German has only second person imperative forms
Table 3.5: Values of morphological features Feature case gender number mood person tense
3.4.3
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)
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, eg. 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). 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 23
Table 3.6: Node labels Node Labels Description Phrase Node Labels ADJX adjectival phrase ADVX adverbial phrase DP determiner phrase (e.g. gar keine) FX foreign language phrase NX noun phrase PX prepositional phrase VXFIN finite verb phrase VXINF non-finite verb phrase Topological Field Node Labels LV resumptive construction (Linksversetzung) C complementizer field (C-Feld) FKOORD coordination consisting of conjuncts of fields KOORD field for coordinating particles LK left sentence bracket (Linke (Satz-)Klammer) MF middle field (Mittelfeld) MFE middle field between VCE and VC NF final field (Nachfeld) PARORD field for non-coordinating particles VC verb complex (Verbkomplex) VCE verb complex with the split finite verb of Ersatzinfinitiv constructions VF initial field (Vorfeld) FKONJ conjunct consisting of more than one field Root Node Labels DM discourse marker P-SIMPX paratactic construction of simplex clauses R-SIMPX relative clause SIMPX simplex clause
24
Table 3.7: 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 OG-MOD, OPP-MOD, OS-MOD, complements or modifiers PRED-MOD, FOPP-MOD, e.g. V-MOD = modifier of the verb OADVP-MOD, OADJP-MOD, V-MOD, MOD-MOD Edge Labels in Split Coordinations ONK, ODK, OAK, PREDK, second conjunct (K) in OPPK, FOPPK, OADVPK, split coordinations OSK, 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
25
Table 3.8: Labels for proper nouns and named entities Labels
Description Phrase Node Labels proper noun or named entity (additional label) Secondary Edge Label phrase internal relation between two parts of a proper noun
EN-ADD EN
annotated trees in one syntactic unit:5 SIMPX 511 −
−
−
−
VF 510 V−MOD PX 506 −
LK 507 HD
HD NX 500 −
An
MF 508
VXFIN 501 HD
HD
ON
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
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 5
These tree diagrams and all following tree diagrams in this report were generated with the aid of the Negra Annotate tool.
26
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: −
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
sie
−
von
der
Polizei
PPER
VAFIN
PROP
VVPP
VAPP
$(
APPR
ART
NN
dsm
dsm
3pit
−−
np*3
3pks
−−
−−
−−
−−
d
dsf
dsf
3.4.5
5
6
7
worden
HD
,
4
geschlagen
NX 507 −
$,
3
dabei
VXINF 506 HD
HD
VVFIN
2
seien
VXINF 505 HD
−
NN
1
erklärten
PX 514 HD
APPRART
0
Abend
VC 513 OV
8
9
10
11
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
$.
−−
−−
−−
−−
27
12
13
PX 503 −
HD PX 502 −
HD
ADVX 500 HD
NX 501 HD
Hoffentlich
ohne
0
Nebenwirkungen
1
.
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 510 −
−
−
VF 508 ON
NX 509 HD
EN−ADD 505 −
LK 506 HD
NX 500 HD
VXFIN 501 HD
ASB
lädt
− NX 507 −
− VC 502 VPT
NX 503 HD
HD
ADJX 504 HD
ein
:
Tag
der
offenen
NN
VVFIN
PTKVZ
$.
NN
ART
ADJA
NN
nsm
3sis
−−
−−
nsm
gsf
gsf
gsf
0
1
2
3
4
5
Tür
6
7
NX 512
EN−ADD 508 −
EN−ADD 509 −
NX 500 −
Arlington
NX 501 HD
−
NX 505 HD
−
NX 507 −
−
,
R
:
Mark
,
D
:
Jeff
,
Tim
$,
NN
$.
NE
NE
$,
NN
$.
NE
NE
$,
NE
NE
nsn
nsf
npm
−−
−−
nsf
−−
nsm
nsm
−−
npm
−−
nsm
nsm
−−
nsm
nsm
4
5
6
7
28
8
9
10
11
12
Bridges
−
CARD
3
Pellington
NX 506 −
NE
2
1999
NX 504 −
EN−ADD 511 −
NE
1
USA
NX 503 HD
KONJ
EN−ADD 510 −
NE
0
Road
NX 502 HD
KONJ
13
14
15
Robbins
16
SIMPX 512 −
−
−
VF 510 V−MOD
MF 511 ON
ADVX 507
NX 500 HD
Berlin
NX 501 HD
LK 508 HD
HD
−
ADVX 502 HD
ADVX 503 HD
PX 509 −
VXFIN 504 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
NX 505 HD
NE
0
also
PRED
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
29
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
0
− PX 508
refint
−
HD 512
NX 502 HD
NX 503 −
Auswirkungen
1
−
auf
2
3
HD
die
Bereitschaft
4
5
−
MF 509 OA
VC 510 HD
NX 504 HD
VXINF 505 HD
,
Therapieangebote
6
7
anzunehmen
8
.
9
PDS
VAFIN
NN
APPR
ART
NN
$,
NN
VVIZU
$.
nsm
3sis
apf
a
asf
asf
−−
apn
−−
−−
−
SIMPX 512 −
−
− NF 511 OS SIMPX 510 −
VF 505 OA
LK 506 HD refcontr
NX 500 −
VXFIN 501 HD
HD
MF 507 ON
MF 508 OD
VC 509 HD 500
NX 502 HD
NX 503
VXINF 504 HD
das
zu
schicken
VVFIN
PIS
ART
NN
PTKZU
VVINF
$.
***
asn
3sks
ns*
dp*
dp*
−−
−−
−−
3.4.7
3
4
Angehörigen
−
PDS
2
den
HD
All
1
man
−
PIDAT
0
versuche
−
5
6
7
.
8
Empty Categories
In general, an empty category analysis, e.g. for phrases without heads, is being avoided in the T¨ uBa-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 labeled with grammatical functions. Furthermore, the edges below the SIMPX node are empty. They are not labeled 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.
30
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
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.
31
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. 8.1.2)). 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
Taten
2
3
NX 501 −
HD
NX 500 HD
Bremens
Gesundheitssenatorin 0
1
NE
NN
gsn
nsf
32
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
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 stategy is justified by the fact that these modifiers have a scope of modification beyond the adjectival phrase, e.g. as in coordinate noun phrases like insgesamt 12.000 Studienpl¨atze und 15.000 Lehrstellen, the adverb insgesamt modifies 12.000 Studienpl¨atze as well as 15.000 Lehrstellen.
33
NX 502 −
−
ADJX 500
HD
ADJX 501
HD
HD
lieber
knieartiger
Leser
0
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. NX 504 −
HD EN−ADD 503 − NX 502 −
HD
EN−ADD 501 − NX 500 −
Die
" 0
−
debis
1
Systemhaus 2
GmbH 3
" 4
5
ART
$(
NE
NE
NN
$(
nsf
−−
nsf
nsn
nsf
−−
NX 503 −
HD EN−ADD 502 − NX 501 −
HD
NX 500 HD
der
Heinrich
ART
NE
NN
gsf
gsm
gsf
0
1
Böll−Stiftung
2
34
NX 504 −
HD
NX 503 HD
− PX 502 −
HD
NX 500 HD
NX 501 HD
"
Solidarität
$(
NN
APPR
−−
nsf
d
0
mit
1
Miloevic
2
"
−
Parolen
NE
$(
$(
NN
dsm
−−
−−
dpf
3
4
5
6
Milosevic 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
4.2.2
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 99 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
35
NX 503 HD
− NX 502 −
−
NX 500
ADJX 501
−
HD
Die
HD
HD
Mitteilung 0
des 1
Bremer
Senats
2
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 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:
36
SIMPX 510 −
−
VF 509 ON NX 508 HD
− PX 507 −
HD NX 506 HD
−
NX 504 − NX 500 −
Der
HD
ADJX 501 HD
HD
Etat
0
LK 505 HD
von
1
3,5
2
NX 502 HD
Millionen
3
4
Mark
VXFIN 503 HD
steht
5
.
6
7
ART
NN
APPR
CARD
NN
NN
VVFIN
$.
nsm
nsm
d
−−
dpf
dpf
3sis
−−
SIMPX 517 −
−
−
VF 516 OS SIMPX 515 −
−
−
VF 513 ON
MF 514 MOD
NX 508 HD
−
LK 509 HD
−
NX 500
NX 501 HD
HD
NX 510 −
VXFIN 502 HD
OA
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
4
zuletzt
VXFIN 505 HD
ART
2
kostete
ADJX 504 HD
MF 512 ON
"
1
Weißbrot
ADVX 503 HD
LK 511 HD
HD
$(
0
Kilo
OA
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. 4.2.3) and complex proper nouns (cf. 4.2.5). NX 503 HD
−
EN−ADD 502 − NX 500 −
Heinz
NX 501 HD
−
(
SPD
NE
NE
$(
NE
$(
nsm
nsm
−−
nsf
−−
0
Schleußer
1
2
3
)
4
37
NX 503 HD
−
NX 502 −
HD EN−ADD 500 −
NX 501 HD
die
Arbeiterwohlfahrt
ART
NN
NE
nsf
nsf
nsn
0
1
Bremen
2
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
−−
4.2.3
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) Landesvor38
sitzende 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 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 labeled 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
morgen 0
HD
, 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 503 APP
APP EN−ADD 502 −
NX 500
NX 501
HD
−
Landesvorsitzende
−
Ute 0
Wedemeier 1
2
NN
NE
NE
nsf
nsf
nsf
39
NX 506 APP
APP NX 505 HD
EN−ADD 503
NX 504
−
−
NX 500
HD
ADJX 501
−
−
Volker
−
NX 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:
40
NX 506 APP
APP NX 505 APP
APP
NX 503 −
EN−ADD 504
−
HD
−
ADJX 500
NX 501
HD
Die
NX 502
HD
Dortmunder 0
Psychologin
−
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 in brackets which are identical to the preceding proper noun, for example, an actor’s name and role: NX 504 APP
APP
EN−ADD 502
EN−ADD 503
−
−
NX 500 −
NX 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 505 −
HD NX 504 APP
APP EN−ADD 503 −
ADVX 500 HD
Auch
0
NX 501 HD
Bundesumweltminister
NX 502 −
1
Jürgen
−
2
Trittin
ADV
NN
NE
NE
−−
nsm
nsm
nsm
3
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.
41
PX 504 −
HD NX 503 APP
APP EN−ADD 502 −
NCX 500 −
HD
in
dem
APPR
ART
d
dsm
0
NCX 501
1
Film
−
HD
"
Das
NN
$(
ART
NN
$(
dsm
−−
nsn
nsn
−−
2
3
Verhör
4
"
5
6
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
−−
−−
−−
−−
−−
−−
−−
4.2.4
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 (4.2.5) 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.
42
DM 500 −
hello 0
FM −−
NX 501 −
−
FX 500 −
−
das
no
0
HD
longer
1
−
amused
2
3
Kollegium
ART
FM
FM
FM
NN
nsn
−−
−−
−−
nsn
4
FX 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).
43
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 Uberschrift ’user als looser’.. 44
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
4.2.5
1
2
3
4
als
HD
5
looser
6
’
7
Proper Nouns and Named Entities
Proper nouns denote individual living beings, objects, 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 (first and last names of persons, names of companies, and geographical names), we also define names of streets and places, individual names of institutions (e.g. Max-Planck-Institut, Deutsches Museum, events (e.g. Zweiter Weltkrieg), and titles of books, movies, etc. as specific categories of proper nouns. Since the PoS-tagging of proper nouns follows the categories of proper nouns in the STTS, some nouns which belong to our class of proper nouns (e.g. composed forms of NE + NN like S¨ ogestraße) are tagged as NN. 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) st¨ urmische (ADJA) Art (NN)). This kind of proper nouns, we define as named entities. In order to distinguish nouns from proper nouns and named entities, the latter ones are assigned an additional node label EN-ADD above their mother node. EN-ADD subsumes all single proper nouns which are not tagged as NE, complex proper nouns tagged as NE as well as named entities tagged according to their distribution. The secondary edge label EN is used for complex proper nouns if EN-ADD cannot be used because of the annotation rules for the internal structure of nouns phrases (e.g. den gleichgeschalteten Hamburger Reichssender ) (cf. 4.2.5). Since German and foreign language named entities differ in terms of their PoS-tagging, their internal syntactic structure also differs in consequence. Our annotation strategy of proper nouns and named entities will be demonstrated in the following. 45
German Proper Nouns German proper nouns denoting individual entities in the above mentioned sense consist of one or more lexical elements tagged as NE. In case of a single NE, this NE is projected to its phrase level, like nouns, carrying the node label NX. Proper nouns 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, e.g. first name and last name of a person, initials, (included) nick names, and names of institutions consisting of NEs. As mentioned above, these complex proper nouns are assigned the additional node label EN-ADD. NX 500 HD
Hamburg
0
NE dsn
EN−ADD 501 − NX 500 −
−
Ute
Wedemeier 0
1
NE
NE
nsf
nsf
EN−ADD 501 − NX 500 −
−
K.
W. 0
1
NE
NE
ns*
ns*
EN−ADD 501 − NX 500 −
−
Ulrich
" 0
−
Tofu
1
" 2
Reineking
3
4
NE
$(
NE
$(
NE
nsm
−−
nsm
−−
nsm
46
EN−ADD 501 − NX 500 −
−
Bayern
München
0
NE
NE
asn
asn
1
Proper nouns which are not tagged as NE, e.g. composed forms of NE+NN (S¨ ogestraße) or complex phrases (Der Spiegel) also get an additional EN-ADD node. If a preceding article does not belong to the proper noun itself, EN-ADD is directly projected from the lexical level of the noun: SIMPX 507 −
−
MF 506 ON
MOD PX 505 −
HD NX 503 −
NX 500 HD
CDU−Treff
in 0
der 1
VC 504 HD
HD
EN−ADD 501
VXINF 502
−
HD
Sögestraße 2
eröffnet 3
4
NN
APPR
ART
NN
VVPP
nsm
d
dsf
dsf
−−
If the proper noun, is a complex syntactic structure, e.g. a phrase or a sentence, the lexical elements of this syntactic structure are tagged according to their distribution. First the whole phrase/sentence is annotated before EN-ADD is added: EN−ADD 501 − NX 500 −
HD
der
Spiegel 0
1
ART
NN
nsm
nsm
If proper nouns include other parts of speech than NEs, these are tagged according to their distribution. Therefore, proper nouns with a preposition include a prepositional phrase.
47
EN−ADD 503 − NX 502 −
− PX 501 −
HD NX 500 HD
Ole
von
Beust
0
1
2
NE
APPR
NE
nsm
d
dsm
If a proper noun occurs within a more complex proper noun, EN-ADD is annotated on both levels of proper nouns: NX 504 −
HD EN−ADD 503 − NX 502 −
HD
EN−ADD 501 − NX 500 −
Die
" 0
−
debis
1
Systemhaus
GmbH
2
3
" 4
5
ART
$(
NE
NE
NN
$(
nsf
−−
nsf
nsn
nsf
−−
If the original form of a proper noun (e.g. Zweiter Weltkrieg, Hamburger Reichssender) is inflected and/or premodified by an article and/or attributive adjective, the included proper noun is indicated by the secondary edge label EN. EN always points from the dependend part of the proper noun to its head noun: SIMPX 509 −
−
− MF 508 PRED NX 507 HD
VF 504
LK 505
ON
HD
NX 500 HD
−
VXFIN 501 HD
Er
− 5
NX 502
HD
Zeitzeuge 1
HD
ADJX 503
HD
ist 0
− NX 506
des 2
EN
Zweiten 3
Weltkrieges 4
. 5
6
PPER
VAFIN
NN
ART
ADJA
NN
$.
nsm3
3sis
nsm
gsm
gsm
gsm
−−
48
NX 502 −
−
− 3
ADJX 500
ADJX 501
HD
den
HD
HD
gleichgeschalteten
EN
Hamburger
0
Reichssender
1
2
3
ART
ADJA
ADJA
NN
asm
asm
***
asm
The secondary edge label is also used to mark that a postmodifier is part of a proper noun: PX 505 −
HD NX 504 HD
− 2
NX 502 −
PX 503 HD
−
ADJX 500
HD NX 501
EN
HD
vom
HD
Hamburger 0
Institut 1
für 2
Sozialforschung 3
4
APPRART
ADJA
NN
APPR
NN
dsn
***
dsn
a
asf
Proper nouns may occur as a prenominal genitive with an attributive function: NX 501 −
HD
NX 500 HD
Bremens
Häfensenator 0
1
NE
NN
gsn
nsm
Proper nouns may be modified postnominally by a another proper noun, e.g. Arbeiterwohlfahrt Bremen (cp. 4.2.2), which is no proper noun as a whole because Bremen only specifies a specific location of die Arbeiterwohlfahrt. Whereas, M¨ unchen in Bayern M¨ unchen is part of the name of the soccer club. Furthermore, Bremen in a construction like das Motorschiff Bremen belongs to an appositional construction (cf. 4.2.3).
Foreign Language Proper Nouns Since all elements of foreign language proper nouns are tagged as NE, they are annotated in the same way as German proper nouns, i.e., they are either a single proper noun or a complex proper noun with an EN-ADD node. Single foreign language proper nouns are projected to their phrase level, complex ones are attached on the same level without head assignment. Their mother node is NX with the additional node EN-ADD. Note the difference between Ole von Beust (cf. 4.2.5) and Inez van Lambsweerde which results from the PoS-tagging rules. If the proper noun 49
consists of more than one lexical element and if it has a German article, the article is attached on a higher level. NX 500 HD
Mary 0
NE dsf
EN−ADD 501 − NX 500 −
−
New
York 0
1
NE
NE
dsn
dsn
EN−ADD 501 − NX 500 −
−
Inez
−
van 0
Lambsweerde 1
2
NE
NE
NE
dsf
dsf
dsf
EN−ADD 501 − NX 500 −
−
University
−
of 0
Illinois 1
2
NE
NE
NE
dsf
dsn
dsn
NX 502 −
HD EN−ADD 501 − NX 500 −
die
−
Tour 0
−
de 1
France 2
3
ART
NE
NE
NE
asf
asf
asf
asf
50
German Named Entities As mentioned above, German named entities are always tagged according to their distribution and annotated with their internal syntactic structure as noun phrases, prepositional phrases, clauses, etc. In order to indicate their status as named entities, the additional node EN-ADD is inserted between their mother node and the next higher annotation level. If two EN-ADD nodes are coordinated, their mother node is NX which represents the nominal status of EN-ADD: SIMPX 518 −
−
−
VF 517 V−MOD PX 516 −
HD EN−ADD 514
MF 515
−
ON
ADJX 512 HD
−
KONJ
PX 507
LK 508
−
HD
ADJX 500
"
Schlaflos
in 2
−
−
NX 503
HD
Seattle 3
" 4
KONJ EN−ADD 510
VXFIN 502
HD
−
EN−ADD 509
HD
NX 501
HD
Seit
PRED
NX 513
gelten
−
Tom 6
HD
NX 504
−
5
NX 511 −
−
Hanks 7
und 8
−
Meg 9
−
NX 505
NX 506
HD
Ryan 10
als 11
−
Dream−Team 12
HD
des 13
Biedersinns
0
1
APPR
$(
ADJD
APPR
NE
$(
VVFIN
NE
NE
KON
NE
NE
KOKOM
NN
ART
14
NN
d
−−
−−
d
dsn
−−
3pis
nsm
nsm
−−
nsf
nsf
−−
nsn
gsm
gsm
If EN-ADD has a premodifier, which can be an EN-ADD itself or a postmodifier, its mother node is also always NX: SIMPX 515 −
−
−
VF 514 ON NX 512 −
MF 513 −
HD
V−MOD
EN−ADD 510 − EN−ADD 506 −
NX 500 −
Bukowskis 0
" 2
Bis
−
VXFIN 503
HD
derbes 1
NX 509
HD
NX 502
HD
HD
LK 508 HD
ADJX 501 −
Oliver
−
PX 507
−
OA
PX 511
ADJX 504
HD
Denver
" 5
NX 505
HD
feierte
6
HD
im
Altonaer 8
Theater 9
Premiere
3
4
NE
NE
ADJA
$(
APPR
NE
$(
VVFIN
APPRART
ADJA
NN
NN
gsm
gsm
nsn
−−
a
asn
−−
3sit
dsn
***
dsn
asf
51
7
HD
10
11
15
NX 510 HD
−
EN−ADD 509 − SIMPX 507
PX 508
−
−
LK 504 HD
ON
VXFIN 500 HD
"
−
HD
MF 505
Sind
EN−ADD 506 PRED
−
NX 501
NX 502
HD
HD
Sie
NX 503 −
Luigi
? 3
"
4
von
−
Stephan
Brüggenthies
0
1
2
5
6
$(
VAFIN
PPER
NE
$.
$(
APPR
NE
7
NE
8
−−
3pis
np*3
nsm
−−
−−
d
dsm
dsm
SIMPX 519 −
−
−
−
VF 518 ON NX 517 HD
−
EN−ADD 516 − SIMPX 514 −
−
−
FOPP
VF 508
LK 509
MF 510
LK 511
ON
HD
OA
HD
NX 500
VXFIN 501
HD
"
MF 515
OA
0
NX 502
HD
HD
jagt 1
Oberärztin 2
" 3
−
ADVX 503
VXFIN 504
HD
HD
jedenfalls
4
VC 513 HD
OV
NX 505 HD
wird 5
PRED
PX 512
vom 6
ZDF 7
NX 506 −
als 8
VXINF 507 HD
" 9
HD
Medicomödie
10
" 11
angepriesen
12
13
$(
NN
VVFIN
NN
$(
ADV
VAFIN
APPRART
NE
KOKOM
$(
NN
$(
VVPP
−−
nsm
3sis
asf
−−
−−
3sis
dsn
dsn
−−
−−
nsf
−−
−−
EN-ADDs which are directly attached to a field are assigned the grammatical function they have in the German sentence: SIMPX 508 −
−
−
− MF 507 PRED
KOORD 500
VF 504
LK 505
ON
HD
NX 501
−
HD
Oder
−
VXFIN 502
NX 503
HD
er 0
EN−ADD 506
−
heißt 1
−
Elvis 2
Costello 3
. 4
5
KON
PPER
VVFIN
NE
NE
$.
−−
nsm3
3sis
nsm
nsm
−−
If named entities, e.g. a title, consist of two separate phrases and/or clauses, (e.g. title and subtitle), the first part is annotated as the head of the second part to express their 52
14
dependency relation: NX 510 APP
APP EN−ADD 509 − NX 508 HD
− NX 507 HD
− PX 506 −
HD
NX 504 −
−
NX 500
−
ADJX 501
−
HD
Die
NX 505 HD NX 502
HD
Ausstellung 0
" 1
Der
2
HD
neue 3
Mensch 4
− 5
HD
Obsessionen
6
HD
ADJX 503
im 7
20. 8
Jahrhundert 9
" 10
11
ART
NN
$(
ART
ADJA
NN
$(
NN
APPRART
ADJA
NN
$(
nsf
nsf
−−
nsm
nsm
nsm
−−
npf
dsn
dsn
dsn
−−
Foreign Language Named Entities The syntactic annotation of foreign language named entities differs from the annotation of German named entities in the following aspects. As mentioned above, foreign language proper nouns are tagged as NE, while all other lexical entries of a foreign language are tagged as foreign language material (FM). A foreign language named entity which consists only of a proper noun, e.g. the title of a movie (Forrest Gumpp) is assigned an EN-ADD label. If a foreign language named entity consists of only FM tagged elements, these elements are directly attached on the same level without internal syntactic structure. Its mother node is marked as FX, e.g. Knockin’ on Heaven’s Door. If a foreign language named entity consists of NE as well as FM tagged elements, e.g. Shakespeare (NE) in (FM) Love (FM), first the annotation stategy described in 4.2.4 is applied. Then, in a second step, the same strategy as for German named entities is applied: the insertion of the EN-ADD node. EN−ADD 501 − NX 500 −
−
Forrest
Gump 0
1
NE
NE
nsm
nsm
53
EN−ADD 501 − FX 500 −
−
Knockin’
−
on 0
−
Heaven’s 1
Door 2
3
FM
FM
FM
FM
−−
−−
−−
−−
EN−ADD 502 − FX 501 −
−
−
NX 500 HD
Shakespeare
in 0
Love 1
2
NE
FM
FM
nsm
−−
−−
4.2.6
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
−−
54
4.2.7
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. 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
55
Complex time expressions or results of competitions are also treated as cardinal numbers: 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
−−
4.2.8
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: EN−ADD 502 − NX 500
NX 501
HD
−
R
:
−
Joel
Schumacher
0
1
XY
$.
NE
2
NE
3
−−
−−
nsm
nsm
NX 501 −
HD
NX 500 HD
D−76351
Linkenheim 0
1
XY
NE
−−
dsn
56
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
NN
CARD
XY
ART
NN
nsm
−−
−−
gsn
gsn
4.2.9
3
4
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), (P¨ utz 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 G¨ astehaus in the preceding sentence: Die italienische Regierung hat die Familie im staatlichen G¨ astehaus 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 unpersonal or agentless constructions such as Es gibt so eine Buchung or Es geht um popul¨ are Unterhaltung. Some verbs optionally permit an expletive subject but also occur with referential 57
subjects such as Max/Es kopft an der T¨ ur. 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 T¨ uBa-D/Z formal subjects and objects are treated like referential pronouns and are labeled 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. Natrlich gibt es so eine Buchung versus *Natrlich gibt so eine Buchung. 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
2
3
Buchung
.
4
5
"
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 labeled 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
58
7
8
9
10
11
eine
HD
PPER
1
erst
OA
KON
0
bekäme
− MF 516 MOD
12
Identität
13
.
14
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 V¨olker zahlen. Vorfeld-es is labeled ES to indicate its purely structural function. In the first release of T¨ uBa-D/Z, 12/2003, Vorfeld-es was integrated by means of ON-MOD. SIMPX 516 −
−
−
− NF 515 ON−MOD R−SIMPX 514 −
VF 508 ES
LK 509
MF 510
HD
NX 500 HD
VXFIN 501 HD
es
MF 512
ON
ON
MOD
NX 502
NX 503
NX 504
ADJX 505
−
ihn 1
C 511
−
OA
HD
zahlen 0
−
HD
die 2
−
Völker 3
, 4
HD
deren
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
Table 4.1 summarizes tests and labels for the different uses of es. Table 4.1: Types of es type test substitutable by other pronouns optional correlates with clausal argument ungrammatical in Mittelfeld edge label
referential formal pronoun argument yes no
correlate no
Vorfeld-es no
no
no
yes
no
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 auer, is analyzed as a copula construction. 59
11
12
SIMPX 523 KONJ
KONJ SIMPX 522 −
−
−
− NF 521 PRED
SIMPX 519 −
−
VF 510
LK 511
ES
HD
V−MOD
VXFIN 501
ADVX 502
HD
HD
NX 500 HD
"
SIMPX 520
−
Es
0
−
MF 512
geschieht
hier
1
VF 513
LK 514
ON
HD
MOD
VXFIN 505
ADVX 506
HD
HD
NX 503
NX 504
HD
HD
nichts 3
, 4
es
5
sei 6
MF 515
VF 516
LK 517
MF 518
ON
HD
OA
NX 507 HD
denn 7
, 8
−
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
ON
−
Determiner Phrases
Certain pronouns serving as determiners in noun phrases may be premodified, for in¨ stance, by degree adverbs such as in so viele Altere, gar kein Schutz, etc. ¨ In the case of so viele Altere, 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 Altere 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
60
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 508 HD
− PX 507 −
HD
EN−ADD 505
EN−ADD 506
−
−
NX 503 HD
NX 504 −
−
NX 500
NX 501
HD
HD
HSV
HD
BU 0
HD
ADJX 502
− 1
Bramfelder
2
SV 3
4
NE
NE
APPR
ADJA
NN
nsm
nsn
a
***
asm
Since pronominal adverbs (PROP) are pronominal forms of a prepostional phrase, they are directly projected to PX:
61
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 PX 504 HD
3
dazu
4
VXINF 505 HD
sagen
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 HD
vom
HD
23. 0
HD
ADJX 501
bis 1
25. 2
Juli 3
4
APPRART
ADJA
APPR
ADJA
NN
dsm
dsm
a
asm
asm
62
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 f¨ unfter November are annotated as a coordinate attributive adjective phrase within a simple noun phrase (cf. 6.4.1). Premodification of 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: elliptical contructions in which a preceding adverb does not semantically modify the prepositional phrase. In this case 63
the adverbial phrase is high attached to an additional level of PX. 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
64
4.5
Adjectival Phrases
We distinguish between attributive adjectives on the one hand and adverbial or predicative adjectives respectively on the other hand. Attibutive adjectives are tagged as ADJA (die traditionellen Elemente) or CARD (20.15 Uhr), whereas adverbial or predicative adjectives are tagged as ADJD (das Gewicht ist gut; den betriebswirtschaftlich gnstigeren Standort) or PWAV (wie wirke ich). The annotation of superlative and comparative forms is explained in section 7.1 on page 119. 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 M¨ utterlein 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 −
− LK 504
ON
HD
NX 500 −
HD
Gewicht 0
ADJX 502 HD
ist 1
MF 505 PRED
VXFIN 501 HD
Das
−
VF 503
gut 2
3
ART
NN
VAFIN
ADJD
nsn
nsn
3sis
−−
65
NX 502 −
−
HD
ADJX 501 −
HD
ADJX 500 HD
den
0
betriebswirtschaftlich
1
günstigeren
2
Standort
ART
ADJD
ADJA
NN
asm
−−
asm
asm
SIMPX 509 −
−
−
3
−
VF 508 ON NX 504 −
−
HD
ADJX 500 HD
Der
0
männliche
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:
66
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. F¨ ur 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:
67
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.
68
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
−
HD
in
die
Erde
ART
ADV
APPR
ART
NN
ADJA
NN
nsf
−−
a
asf
asf
nsf
nsf
0
teilweise
NX 501
1
2
3
4
gebaute
5
Sporthalle
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
−−
−−
−−
69
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. SIMPX 517 −
−
−
−
−
MF 516 OD
ON
MOD
V−MOD
NX 515
VF 509
LK 510
V−MOD
HD
KOORD 500
ADVX 501
VXFIN 502
−
HD
HD
Doch
heute 0
APP
APP
EN−ADD 511
EN−ADD 512
−
−
NX 503
NX 504
−
will
−
Ina
1
2
ADJX 513 −
−
Terre 3
( 4
−
Hannelore 5
NX 505 HD
Droege 6
) 7
es 8
VC 514 HD
OV
ADVX 506
ADVX 507
VXINF 508
HD
HD
HD
nicht 9
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
ADV
ADV
CARD
2
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
−−
−−
1
bis zu, u ¨ber are considered to be ADV rather than APPR because of their semantic meaning.
70
14
SIMPX 510 −
−
−
−
MF 509 V−MOD VF 505
LK 506
ON NX 500 −
HD
Der
Fahrer 0
ADVX 507
VC 508
HD
HD
−
OV
VXFIN 501
ADVX 502
ADVX 503
VXINF 504
HD
HD
HD
HD
konnte 1
nicht 2
mehr 3
bremsen 4
. 5
6
ART
NN
VMFIN
PTKNEG
ADV
VVINF
$.
nsm
nsm
3sit
−−
−−
−−
−−
4.7
Verb Phrases
Whereas finite verb phrases are labeled VXFIN, non-finite verb phrases are labeled VXINF. Since infinitives and past participles share certain properties (e.g. exchangeability in Man hat nur noch das eigene Herz schlagen h¨oren/geh¨ ort.), 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.
71
SIMPX 513 −
−
−
−
MF 512 OA
FOPP
VF 510
PX 511
ON
−
NX 506 −
−
HD
NX 508
HD
ADJX 500
NX 502
HD
überehrgeizige
Bürgermeister
0
2
HD
die 3
−
NX 503
−
will
1
VC 509
HD
VXFIN 501
HD
Der
HD
LK 507
−
Bergidylle 4
in 5
HD
ein 6
OV
NX 504 −
Mekka 7
HD
des 8
VXINF 505 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
OV
HD
VXINF 503
VXINF 504
HD
HD
gewesen
2
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
−−
−−
nsn
−−
−−
3sis
72
12
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 h¨atten] [MFE mißtrauisch] [VC machen m¨ ussen]).
73
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
und
0
1
ich 2
dazu 3
VC 509
HD
OV
HD
VXFIN 504
VXINF 505
VXINF 506
HD
HD
HD
haben 4
beitragen 5
können 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
2
mißtrauisch
machen
3
4
müssen
5
In the example below, the finite verb precedes the non-finite verbs although m¨ ussen 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
ART
NE
PRF
ADV
ADV
ART
ADV
ADJA
NN
VAFIN
VVINF
VMINF
−−
nsf
nsf
ds*3
−−
−−
asn
−−
asn
asn
3sis
−−
−−
74
7
Konzept
0
KOUS
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 −
LK 507 HD
HD
NX 500 HD
NX 501 KONJ
Erkenntnisse
−
zu
sein
TRUNC
KON
NN
VVFIN
PPER
ADJD
PTKZU
VAINF
$.
npf
gsf
−−
−−
gsf
3pis
dsm3
−−
−−
−−
−−
4
5
6
fremd
−
Friedens−
3
ihm
VXINF 505 HD
der
2
scheinen
ADJX 504 HD
ART
1
Konfliktforschung
NX 503 HD
VC 509 OV
NN
0
und
OD
VXFIN 502 HD
KONJ
MF 508 PRED
7
8
9
.
10
SIMPX 510 −
−
−
−
VF 509 OPP PX 505 −
LK 506 HD NX 500 HD
Über
Details 0
MF 507
HD
MOD
VXFIN 501
ADVX 502
HD
HD
werde 1
VC 508 OV VXINF 503 HD
noch 2
HD
zu 3
VXINF 504 −
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:
75
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 1
Baugesetzen 2
−
auch 3
VXINF 504 HD
mehr 4
HD
Mitspracherechte 5
einzufordern
.
6
7
8
KOUI
APPR
ADJA
NN
ADV
PIAT
NN
VVIZU
$.
−−
d
dpn
dpn
−−
***
apn
−−
−−
Besides 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
VC 509
MF 510
VC 511
HD
OA
HD
VXFIN 505
NX 506
HD
Lebensaufgabe 4
−
−
machen 5
, 6
VXINF 507 HD
den
HD
Kranken
7
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
ON
VXINF 502 HD
C 503 −
zu 1
−
−
MF 509
−
sagen 2
, 3
daß 5
HD
NX 505 HD
ihr
4
OA
NX 504 −
VC 510
Zug 6
VXFIN 506
HD
HD
Verspätung 7
hat 8
9
KOUI
NN
PTKZU
VVINF
$,
KOUS
PPOSAT
NN
NN
VAFIN
−−
dpm
−−
−−
−−
−−
nsm
nsm
asf
3sis
76
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
V−MOD
LK 508
ON
HD
NX 500
−
wendet 0
HD
NX 502
HD
Er
LK 510
−
VXFIN 501
HD
−
PX 509
den 1
HD
NX 503 HD
−
Blick 2
von 3
HD
VXFIN 504 HD
der 4
VC 511
HD
Wand 5
VC 505
und 6
VPT
fängt 7
HD
an 8
VXINF 506 −
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, m¨ochten 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. u ¨berreden, u ¨berzeugen incoherent, extraposition obligatory: a. [wie er sie u ¨berredet,][mit kritischen politischen Gegenpositionen umzugehen] b.*[wie er sie [mit kritischen politischen Gegenpositionen umzugehen] u ¨berredet]
77
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 u ¨berredet, 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, h¨oren, f¨ uhlen, sp¨ uren) which subcategorize for an accusative and an infinitive. The verbs lassen, machen, heien 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 h¨ort das eigene Herz schlagen). The infinitive as the verbal 78
object of the AcI verb is located in the verb complex and the accusative is realized as OA in MF. SIMPX 510 −
−
−
MF 509 ON
MOD
MOD
OA NX 507 −
C 500
NX 501
−
HD
Wenn
man 0
ADVX 502
ADVX 503
HD
HD
nur 1
−
HD
ADJX 504 HD
noch 2
VC 508
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 labeled OA. SIMPX 509 −
− LK 506 HD VXFIN 500 HD
Lassen
OA
NX 501 HD
NX 502 HD
uns
VVFIN
PPER
3pis
np*3
4.7.7
MF 507 OA
ON
Sie
0
1
−
MOD
NX 503 −
VC 508 OV
HD
ADVX 504 HD
VXINF 505 HD
jetzt
beenden
die
Diskussion
PPER
ART
NN
ADV
VVINF
ap*1
asf
asf
−−
−−
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:2sip)
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:3pip) (warten/VVINF:–) 79
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
SIMPX 506 −
−
VF 503
−
LK 504
V−MOD
MF 505
HD
PX 500
V−MOD
VXFIN 501
HD
ADJX 502
HD
Drum
HD
prüfe
ewig
0
1
2
PROP
VVIMP
ADJD
−−
s
−−
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
−−
SIMPX 507 −
−
− MF 506 ON
VF 503 V−MOD
LK 504 HD
ADJX 500 HD
VXFIN 501 HD
−
lebe
Ned
Lang
0
1
EN−ADD 505 − NX 502 −
2
Devine
ADJD
VVIMP
NE
NE
−−
s
nsm
nsm
3
80
SIMPX 515 −
−
− NF 514 OS SIMPX 513 −
−
−
LK 507 HD
MF 508 ON
VF 509 ON
LK 510 HD
VXFIN 500 HD
NX 501 HD
NX 502 HD
VXFIN 503 HD
−
habe
den
Sage
niemand
0
1
,
das
2
3
−
MF 511
4
VC 512 OV
OA
MOD
NX 504
ADVX 505 HD
VXINF 506 HD
nicht
verändert
HD
Westen
5
6
7
.
8
9
VVIMP
PIS
$,
PDS
VAFIN
ART
NN
PTKNEG
VVPP
$.
s
ns*
−−
nsn
3sks
asm
asm
−−
−−
−−
4.7.8
Particle Verbs
Separable verb particles are tagged as PTKVZ and annotated with the edge label VPT: SIMPX 511 −
−
VF 509
MF 510 OD LK 507
HD
ADVX 500
−
NX 501
HD
−
Auch
HD
der 2
−
−
VXFIN 503
−
Vertreter 1
NX 508
HD
NX 502 HD
die 0
−
ON 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
−−
−−
In verb-final clauses, the particle verb occurs unseparated within the verb complex: SIMPX 510 −
−
VF 506
−
LK 507
ON NX 500 HD
Rußland
−
MF 508
VC 509
HD
MOD
OD
MOD
OV
VXFIN 501
ADVX 502
NX 503
ADVX 504
VXINF 505
HD
HD
HD
HD
wollte 0
−
bislang 1
HD
einer 2
UN−Resolution 3
nur 4
zustimmen 5
6
NE
VMFIN
ADV
ART
NN
ADV
VVINF
nsn
3sit
−−
dsf
dsf
−−
−−
81
4.7.9
Verbs with Predicate
Typically, the complement type PRED (predicate) occurs with verbs like sein, haben, scheinen, aussehen, sich anh¨ oren, 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 labeled as PRED (e.g. Unter dem Motto Kino-Extrem agiert der Regisseur als Filmjockey.) SIMPX 511 −
−
−
VF 510 V−MOD PX 509 −
HD NX 508 APP
APP EN−ADD 505
LK 506
− NX 500 −
Unter
dem 0
NX 501 HD
−
agiert 3
PRED
NX 503
HD
Kino−Extrem 2
ON
VXFIN 502
HD
Motto 1
MF 507
HD
HD
der 4
NX 504 −
Regisseur 5
HD
als 6
Filmjockey 7
8
APPR
ART
NN
NN
VVFIN
ART
NN
KOKOM
NN
d
dsn
dsn
dsn
3sis
nsm
nsm
−−
nsm
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:
82
SIMPX 508 −
−
−
VF 504
LK 505
ON
HD
NX 500
MF 506
ADVX 503
HD
ist 0
V−MOD
NX 502
HD
Das
NF 507
PRED
VXFIN 501
HD
−
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
MF 506
VXFIN 501
HD
HD NX 503
HD
ist 0
−
ADJX 502
HD
es
PX 507
PRED
−
kalt 1
an 2
HD
diesem
Tag
3
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. R−SIMPX 511 −
−
−
MF 510 V−MOD C 505 ON
PRED
PX 506 −
NX 507 HD
NX 500 −
das
−
an 0
HD
−
der 3
HD
schönste 4
HD
NX 503
HD
Abend 2
VC 509 HD
ADJX 502 HD
diesem 1
PX 508
−
NX 501
HD
MOD
Platz 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):
83
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
HD
Freunde
NX 504
HD
wurde
1
HD
ADJX 503
2
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
MF 506 PRED
VXFIN 501
HD
ADJX 502
HD
ich
FOPP PX 503
HD
bin
HD
froh
0
1
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: SIMPX 511 −
−
−
−
VF 510 ON EN−ADD 506
LK 507
−
MF 508
HD
NX 500
PRED
VXFIN 501
HD
NX 502
HD
Bravo
OV
HD
ADVX 503
VXINF 504
VXINF 505
HD
HD
HD
HD
kann 0
VC 509 MOD
es 1
nicht 2
gewesen 3
sein 4
. 5
6
NN
VMFIN
PPER
PTKNEG
VAPP
VAINF
$.
nsf
3sis
nsn3
−−
−−
−−
−−
SIMPX 507 −
−
VF 504 ON
−
LK 505 HD
NX 500 HD
PRED
VXFIN 501 HD
er
NX 502 HD
war 0
MF 506
HD
es 1
MOD ADVX 503
nicht 2
3
PPER
VAFIN
PPER
PTKNEG
nsm3
3sit
nsn3
−−
84
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
VXFIN 501
HD
"
−
Die
0
NX 502
HD
VXINF 503
−
wollten 1
HD
die 2
HD
BLG 3
schonen 4
. 5
"
6
7
$(
PDS
VMFIN
ART
NE
VVINF
$.
$(
−−
np*
3pit
asf
asf
−−
−−
−−
SIMPX 512 −
−
LK 509
−
MF 510
HD VXFIN 500 HD
Hätte
ON
OD
OA
MOD
NX 501
NX 502
NX 503
ADVX 504
HD
HD
HD
sie 0
VC 511
sich
HD
das
1
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
NX 501 HD
0
Daewoo
VC 506 HD
HD NX 502 HD
1
nach
2
Bremen
VXFIN 503 HD
3
mußte
PWAV
NE
APPR
NE
VMFIN
−−
ns*
d
dsn
3sit
4
85
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 ON
VXFIN 502 HD
der 2
LK 505 HD
Menge 3
NX 503
HD
−
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:
86
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: 87
SIMPX 511 −
−
−
−
VF 510 OA−MOD PX 506
LK 507
−
HD
HD
NX 500
VXFIN 501
−
Für
HD
diese 0
MF 508
HD
Behauptung 1
ON
MOD
OA
NX 502
ADVX 503
NX 504
HD
hat 2
VC 509
HD
Beckmeyer 3
OV
−
bisher 4
HD
keinen 5
VXINF 505 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
MF 512
V−MOD
ON
ADJX 507 KONJ
−
KONJ
MOD
HD
HD
energisch 1
NX 503 HD
will 2
VC 510
−
VXFIN 502
HD
und
PX 509
HD
ADJX 501
0
OA
LK 508
ADJX 500
Phantasievoll
−
−
er 3
HD
NX 504 HD
den 4
OV
NX 505 −
Entwicklungsprozeß 5
an 6
HD
der 7
VXINF 506 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 g¨abe (intelligente L¨osungen OA), (die kein Geld kosten OA-MOD)). 88
Typical MODs and V-MODs: Generally, modifying subclauses (e.g. Katastrophenstimmung herrscht erst, [wenn 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, nat¨ urlich 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, daf¨ ur, 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, sp¨ater, 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 attachement) 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
ADVX 502 HD
NX 508 −
VXFIN 503 HD
−
HD
ADJX 505 HD
HD
für
Rumänien
der
Papst−Besuch
APPR
NE
ADV
VAFIN
ART
NN
APPR
ADJA
NN
$.
−−
a
asn
−−
3sis
nsm
nsm
d
dsf
dsf
−−
1
5.3.2
2
ist
NX 504
ADV
0
selbst
HD 509
LK 507 HD
3
4
5
von
6
7
großer
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 510 −
−
−
VF 508
NX 509
ON
HD
EN−ADD 505 −
−
LK 506
NX 507
HD
NX 500 HD
−
VXFIN 501 HD
ASB
lädt 0
VC 502
NX 503
VPT
HD
ein 1
: 2
HD
HD
Tag
3
− ADJX 504
der 4
offenen 5
Tür 6
7
NN
VVFIN
PTKVZ
$.
NN
ART
ADJA
NN
nsm
3sis
−−
−−
nsm
gsf
gsf
gsf
89
NX 502 − NX 500
ADJX 501
−
HD
Sein
HD
HD
Zuhause
:
0
1
stilvolles
Entertainment
2
.
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 ON NX 500
VF 507
HD
ON
VXFIN 501
HD
meinte
: 1
−
MF 509
HD
PRED
VXFIN 503
HD
0
−
LK 508
NX 502
HD
Er
−
LK 506
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
NX 506 −
auch 8
−
in 9
der 10
NX 507 HD
−
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. 90
The overall attachment strategy has been chosen in order to keep syntactic structure 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
−−
91
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
VC 507
MOD
ON
ADVX 501
NX 502
OV refvc
C 500 −
HD
Wenn
HD
da 0
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.
92
SIMPX 510 −
−
−
MF 509 ON
V−MOD
MOD
PRED
PX 508 −
HD NX 506
VC 507
− C 500
NX 501
−
−
so
−
daß 0
HD
der 1
ADVX 502 HD
Maschinenpark 2
HD
HD
ADJX 503
ADJX 504
HD
heute 3
für 4
VXFIN 505
HD
lukrative 5
Sonderanfertigungen 6
HD
unbrauchbar
ist
7
8
9
KOUS
KOUS
ART
NN
ADV
APPR
ADJA
NN
ADJD
VAFIN
−−
−−
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
−
KONJ
−
NX 502
NX 503
NX 504
−
HD
HD
HD
wenn
man 2
wie 3
ARD 4
und 5
HD
DP 510
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.
93
SIMPX 508 −
−
C 505
−
MF 506
ON
MOD
NX 500
ADVX 501
HD
PRED
ADJX 502
HD
Wieviel
VC 507
V−MOD
ADJX 503
HD
da 0
HD VXFIN 504
HD
monatlich
HD
fällig
1
2
wird 3
4
PWS
ADV
ADJD
ADJD
VAFIN
nsn
−−
−−
−−
3sis
R−SIMPX 510 −
−
C 508
MF 509
FOPP
ON
PX 505
MOD
MOD
NX 506
−
HD
−
HD
HD
ADJX 501 HD
deren 0
VC 507
−
NX 500 −
zu
−
HD
Ablaß 1
die
tonale
2
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. H¨ohle (1986)). Here are two examples of different clause types: 94
−
−
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.
95
6.1.5
Resumptive Constructions: The LV-Field
Resumptive constructions are analyzed as suggested by H¨ohle (1986) and Kathol (1995), by using the field LV (Linksversetzung) which is located on the left side 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
HD
introvertierten
Einzelgänger 1
zum 2
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
LK 513 HD
NX 502
HD
0
VF 512 ON
Popstar 4
, 5
das
6
MF 514
muß 7
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. 96
15
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): SIMPX 519 −
−
−
−
LV 518 MOD−MOD SIMPX 516 −
MF 517
−
−
MF 511 MOD
ON
ADVX 501
NX 502
OV
−
HD
Wenn
HD
da 0
was 1
VF 513
refmod HD
MOD 516
HD
VXINF 504
VXFIN 505
ADVX 506
VXFIN 507
HD
HD
HD
HD
HD
gebucht
OV 503
worden
2
ist
3
4
, 5
dann
6
MOD
PRED
LK 514
VXINF 503
refvc C 500
ON
VC 512
− NX 508 HD
ist 7
PX 515
NX 510
HD
das 8
HD
ADVX 509
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 labeled 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, wofr, 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. The annotation of wie is still work in progress. In the current release it is annotated as ADJX instead of ADVX if it directly modifies an adjective or adverb itself. SIMPX 510 −
−
VF 507 V−MOD
LK 508 HD
PX 500 HD
VXFIN 501 HD
Warum
0
machen
−
−
MF 509 ON
OA
NX 502 HD
1
wir
2
NX 503 −
den
HD
3
Computer
4
MOD
MOD
ADVX 504 HD
ADVX 505 HD
nicht
einfach
5
VC 506 VPT
6
aus
7
?
8
PWAV
VVFIN
PPER
ART
NN
PTKNEG
ADV
PTKVZ
$.
−−
1pis
np*1
asm
asm
−−
−−
−−
−−
97
6.2.2
Yes - No Questions
Yes - no questions may occur in various forms, but the most typical form is the verb-first clause: SIMPX 506 −
− MF 505 ON
LK 503
OA
NX 504
HD
−
HD
VXFIN 500
EN−ADD 501
HD
−
Veruntreute
die 0
NX 502 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
−−
98
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 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
VF 507
LK 508
ON KOORD 500
HD
NX 501
−
Aber
C 510 HD
HD
gäbe 1
NX 509
ADJX 503
HD
es 0
−
−
VXFIN 502
HD
R−SIMPX 514
OA
− MF 511
ON
OA
NX 504
NX 505
HD
intelligente 2
Lösungen 3
, 4
−
die
5
− VC 512 HD 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 labeled 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
C 513
OPP VF 506
LK 507
ON
HD
NX 500 −
Ein
PX 508 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
VXFIN 501 HD
−
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 on the right side 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 Gef¨ angnisgitter 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 99
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. 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 516 −
−
−
MF 515 OA NX 514 HD
− R−SIMPX 513 −
− C 507 ON
NX 508 −
NX 500 HD
HD
C 509 V−MOD
EN−ADD 501 −
ADVX 502 HD
−
MF 510 ON
PRED
NX 503 HD
NX 504 HD
VC 511 HD
VC 512 HD
VXFIN 505 HD
VXFIN 506 HD
die
die
AWO
,
wo
er
Kreisvorsitzender
ist
,
prüfte
PRELS
ART
NN
$,
PWAV
PPER
NN
VAFIN
$,
VVFIN
$.
nsf
asf
asf
−−
−−
nsm3
nsm
3sis
−−
3sit
−−
0
1
6.3.1
2
3
4
5
6
7
8
.
9
10
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 Relativs¨atze’) do not modify a head word but substitute an argument or adjunct in the clause. Consequently, they are labeled SIMPX on sentential level (instead of R-SIMPX) and they function as (sentential) subject (ON) or sentential object (OS). The latter is 100
.
15
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. SIMPX 516 −
−
−
VF 515 ON SIMPX 513 −
−
C 508 ON
MOD
NX 500 HD
ADVX 501 HD
Wer
−
MF 509 V−MOD
VXINF 503 HD
VXINF 504 HD
HD
VXFIN 506 HD
NX 507 −
will
,
kommt
VVPP
VAINF
VMFIN
$,
VVFIN
APPR
PPOSAT
NN
$.
ns*
−−
−−
−−
−−
3sis
−−
3sis
a
ap*
ap*
−−
−
SIMPX 517 −
−
3
4
5
6
−
7
auf
8
seine
HD
ADJD
2
werden
VXFIN 505 HD
PX 512 −
ADV
1
unterhalten
LK 511 HD
HD
PWS
0
gut
VC 510 OV refvc 503
OV
ADJX 502 HD
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
6
man
7
8
9
Independent relative clauses introduced by wie are currently annotated in a different manner. Wie is analyzed 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 labeled 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). 101
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, 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
−
Kämpfe 1
NX 503
HD
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
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).
102
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
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:
103
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
u.
5
Di.
6
7
ADV
$,
NN
$,
NN
$,
KON
NN
−−
−−
nsm
−−
nsm
−−
−−
nsm
SIMPX 519 −
−
− MF 518 PRED ADJX 517 KONJ
KONJ
KONJ
PX 515
NX 516
KONJ VF 509
LK 510
ON
HD
NX 500 HD
Die
KONJ
VXFIN 501
−
−
KONJ
und 3
geschmackvoll 4
, 5
von
−
HD
PX 513 HD
HD
zart 2
−
ADJX 503
HD
ist 1
PX 512
ADJX 502
HD
Farbpalette 0
ADJX 511
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
$.
nsf
nsf
3sis
−−
−−
−−
−−
d
dsn
a
asn
−−
−−
nsn
−−
nsn
−−
SIMPX 512 −
−
−
−
MF 511 PRED NX 510 − VF 505 ON
KONJ
LK 506 HD
NX 500 HD
−
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
HD
dem 6
VXINF 504 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:
104
15
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
HD
ADJX 507
HD
obendrein 4
−
ADJX 508
HD
kalt 5
,
sondern
bei 8
HD
20 9
Grad 10
erträglich 11
.
6
7
ADV
VAFIN
PPER
PTKNEG
ADV
ADV
ADJD
$,
KON
APPR
CARD
NN
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
außenpolitisch 5
NX 505
HD
als 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 −
VXFIN 500
NX 501
HD
HD
haben 0
−
ADVX 512
HD
Entweder
SIMPX 521
die 1
VC 513 HD
VXINF 503
HD
HD
überhaupt 2
nicht 3
C 514
OV
ADVX 502
OPP
begriffen 4
−
, 5
−
VC 516
VF 517
LK 518
HD
ON
HD
HD
HD
VXFIN 506
NX 507
HD
es 7
−
MF 515
NX 505
worum
−
ON
PX 504
6
SIMPX 522 −
HD
geht
, 9
oder
10
HD
es 11
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
−−
−−
105
8
VXFIN 508
MF 519 OD
15
16
6.4.4
Coordinations with Truncated Words
In contrast to complete lexical entries, a truncated word is directly attached to the conjunct and the conjunction. Neither the truncated word nor the second conjunct are projected firstly to the phrase level. Their edge label is KONJ. Both conjuncts will only then project to a phrasal node if either both conjuncts comprise a determiner or if both conjuncts are premodified. In this case only the complete lexical entry carries a head label. The truncated word does not receive morphological annotation. NX 500 KONJ
−
Bau−
KONJ
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 3
CDU−Veranstaltung 4
5
APPR
ART
TRUNC
KON
ART
NN
d
dsf
−−
−−
dsf
dsf
NX 502 KONJ
−
KONJ NX 501 −
HD
ADVX 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 by means of a shallow structure in the same manner as in the case of less complex structures. ADJX 501 −
KONJ
−
−
KONJ
ADVX 500 HD
sowohl
kultur− 0
als 1
auch 2
stadtentwicklungspolitisch 3
4
KON
TRUNC
KON
ADV
ADJD
−−
−−
−−
−−
−−
106
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
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 507 HD
− PX 506 −
HD EN−ADD 505 −
NX 503 −
KONJ
NX 504 KONJ
NX 500 HD
−
NX 501 HD
Originaltitel
ADJX 502 HD
"
8
MM
"
.
NN
KON
NN
APPR
$(
CARD
NN
$(
$.
nsm
−−
nsf
d
−−
−−
dsm
−−
−−
6.4.5
0
und
HD
1
−fassung
2
von
3
4
5
6
7
8
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 coordinate part is assigned the head function. NX 503 −
HD NX 502 KONJ
den
−
KONJ
NX 500
NX 501
HD
HD
Angestellten 0
und 1
Beamten 2
3
ART
NN
KON
NN
dp*
dp*
−−
dpm
107
NX 504 −
−
HD NX 503 KONJ
ADJX 500 HD
Die
türkischen 0
−
KONJ
NX 501
NX 502
HD
HD
Instrumente 1
und 2
Harmonien 3
4
ART
ADJA
NN
KON
NN
npn
npn
npn
−−
npf
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 labeled 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 − VF 510 ON
HD
LK 511
HD
−
−
VXFIN 501
HD
HD
glauben 0
LK 513
ADJX 502
HD
Wir
V−MOD
NX 512
HD
NX 500
MF 519
an 1
die 2
totale 3
Gegenwart 4
und 5
OA
−
KONJ ADVX 504
ADVX 505
HD
HD
HD
hier 7
KONJ
und 8
−
C 515
HD
tun
−
ADVX 514
VXFIN 503
6
R−SIMPX 520
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:
108
15
16
SIMPX 520 −
−
− FKOORD 519 KONJ
−
KONJ
FKONJ 518 −
−
MF 517 ON
V−MOD
FOPP
PX 515
FKONJ 516
− VF 508 MOD
LK 509 HD
ADVX 500 HD
VXFIN 501 HD
Nun
sollen
HD
−
NX 510
PX 511
− NX 502 HD
−
HD
ADJX 503 HD
NX 504 HD
VXINF 505 HD
NX 506 −
in
zwei
zu
Hobbypolizisten
APPR
CARD
NN
APPR
NN
VAINF
KON
ART
NN
VVINF
$.
−−
3pis
np*
d
−−
dpf
d
dpm
−−
−−
asf
asf
−−
−−
3
4
5
6
7
und
VXINF 507 HD
HD
NN
2
werden
VC 514 OV
VMFIN
1
Wochen
−
MF 513 OA
ADV
0
Leute
HD
VC 512 OV
8
eine
9
Waffe
10
11
bekommen
.
12
13
A coordination of fields may also be an embedded structure. In this case, FKOORD functions also as conjunct label: SIMPX 521 −
− FKOORD 520 KONJ
KONJ FKONJ 519 −
− FKOORD 518 KONJ
FKONJ 515 − VF 508 ON
HD
NX 500 −
HD
Älteren 0
LK 511 −
VXFIN 503
, 3
HD
HD
haben
4
Verpflichtungen 6
und 7
VC 514
MOD
OA
ADVX 505
NX 506
HD
familiäre 5
−
MF 513
ADJX 504
HD
teurer 2
−
NX 512
HD
ADJX 502 HD
sind 1
KONJ FKONJ 517
OA
MF 510 PRED
VXFIN 501 HD
Die
−
LK 509
−
MF 516
−
oft 8
HD
ein 9
OV VXINF 507 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:
109
ADJX 504 −
HD ADJX 503 KONJ
ADVX 500 HD
−
KONJ
ADJX 501
ADJX 502
HD
Viel
HD
größer
und
0
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. SIMPX 512 −
−
− MF 511 OPP PX 510 KONJ
VF 506
LK 507
ON NX 500
HD
−
VXFIN 501
ADVX 502
HD
HD
HD
Wir
glauben 0
−
KONJ
PX 508 −
HD
−
NX 503
an 2
HD
die 3
−
HD
ADVX 504
−
nicht 1
PX 509
NX 505
HD
Vergangenheit
und
4
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 514 −
−
−
−
VF 513 ON NX 511
VC 512
APP
APP
NX 506 −
OV
EN−ADD 507 HD
−
NX 500
NX 501
HD
−
Radunskis
Sprecher 0
, 1
MF 509 OA
VXFIN 502 −
Axel
2
LK 508 HD
, 4
−
wollte
5
KONJ
NX 503
HD
Wallrabenstein 3
VXINF 510 −
HD
die 6
Entscheidung 7
weder 8
−
KONJ
VXINF 504
VXINF 505
HD
HD
bestätigen 9
noch 10
dementieren 11
12
NE
NN
$,
NE
NE
$,
VMFIN
ART
NN
KON
VVINF
KON
VVINF
gsm
nsm
−−
nsm
nsm
−−
3sit
asf
asf
−−
−−
−−
−−
110
13
SIMPX 514 −
−
−
−
MF 513 OA NX 512 − VF 508 ON
LK 509 HD
NX 500
VXFIN 501 HD
−
Die
HD
hatte
KONJ
−
−
KONJ NX 510 −
− ADVX 502 HD
ADVX 503 HD
nicht
etwa
NX 504 −
ADVX 505 HD
HD
,
sondern
PTKNEG
ADV
PPOSAT
NN
$,
KON
ADV
ART
ADJA
NN
VVPP
$.
nsf
nsf
3sit
−−
−−
asf
asf
−−
−−
−−
asn
asn
asn
−−
−−
3
4
5
6
7
8
gleich
das
9
ganze
VXINF 507 HD
VAFIN
2
Lektüre
HD
ADJX 506 HD
NE
1
unsere
VC 511 OV
ART
0
Lufthansa
−
10
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: 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
MOD
MOD
MOD
PRED
NX 502
ADVX 503
ADVX 504
ADVX 505
ADJX 506
HD
HD
HD
es
nicht 2
noch 3
ADJX 508
HD
kalt 5
HD
ADJX 507
HD
obendrein 4
−
,
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
NX 512
ON
HD
wird 0
HD
MF 511
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:
111
13
.
14
SIMPX 523 KONJ
−
KONJ
SIMPX 522 −
−
− MF 521 OPP
VF 519
PX 520
ON
−
HD
NX 516
NX 517
APP
APP EN−ADD 509
Nashorn−Bürgermeister
−
Henning 0
−
VXFIN 502
−
Storchbein
setzt 2
auf 3
HD
die
HD
ADJX 505
HD
Sammlung
der 6
positiven
Kräfte
und
8
9
prompt 10
−
MF 514
VC 515
HD
MOD
OV
VXFIN 506
ADVX 507
VXINF 508
HD
HD
HD
HD
7
−
LK 513
V−MOD
ADJX 504
−
−
VF 512
−
NX 503
HD
1
−
NX 511
HD
NX 501
HD
−
LK 510
− NX 500
SIMPX 518
HD
wird 11
da 12
gepöbelt 13
.
4
5
NN
NE
NE
VVFIN
APPR
ART
NN
ART
ADJA
NN
KON
ADJD
VAFIN
ADV
VVPP
14
$.
nsm
nsm
nsm
3sis
a
asf
asf
gpf
gpf
gpf
−−
−−
3sis
−−
−−
−−
SIMPX 520
−
KONJ
KONJ
SIMPX 518 −
SIMPX 519 −
VF 514 MOD
ON
ADVX 508 −
ADVX 500 HD
ADVX 501 HD
aber
PX 510 −
VXFIN 502 HD
NX 503 HD
blieb
alles
− MF 517 PRED
NX 511 HD
−
NX 504 HD
LK 512 HD
HD
ADVX 505 HD
NX 513 −
−
VXFIN 506 HD
−
nur
der
Innensenator
ein
neuer
VVFIN
PIS
APPRART
NN
$(
ADV
ART
NN
VAFIN
ART
ADJA
$.
−−
−−
3sit
nsn
dsn
dsn
−−
−−
nsm
nsm
3sis
nsm
nsm
−−
2
3
4
5
6
7
8
ist
ADJX 507 HD
ADV
1
alten
−
ADV
0
beim
−
VF 516 ON
OPP
LK 509 HD
HD
So
−
MF 515
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
HD
haben 3
NX 505
? 4
5
,
−
kontert
6
der 7
PX 514 −
VXFIN 506 HD
HD
Popstar 8
und 9
HD
NX 507 −
wirft 10
OPP
HD
einen 11
ADVX 508 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. 112
15
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
NX 505
−
,
wie
diese
Leistung 6
OV 506
HD
VXINF 506
VXINF 507
VXFIN 508
HD
HD
HD
HD
beurteilt 7
werden 8
−
C 518 ON
kann 9
− 10
und
11
VC 520
OPP
NX 509
PX 510
HD
HD
wer 12
−
MF 519 PRED ADJX 511 HD
dafür 13
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
−−
6.4.9
3
−
−
VC 517
17
18
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 labeled as P-SIMPX instead of SIMPX. P−SIMPX 520 KONJ
−
KONJ SIMPX 519 −
−
−
SIMPX 517 −
−
−
OA
VF 515
NX 516
ON
−
NX 508 KONJ
−
KONJ
NX 500
NX 501
HD
HD
Kilos
und 0
−
MF 518
Fitneß 1
LK 509
VC 510
VF 511
LK 512
HD
OV
ON
HD
VXFIN 502
VXINF 503
HD
HD
sollen 2
NX 504
, 4
denn
5
OV
NX 506 −
will 7
VC 514
−
HD
Wesemann 6
EN−ADD 513
VXFIN 505
HD
stimmen 3
HD
die 8
−
Tour 9
de 10
VXINF 507 −
HD
France 11
gewinnen 12
. 13
14
NN
KON
NN
VMFIN
VVINF
$,
KON
NE
VMFIN
ART
NE
NE
NE
VVINF
$.
npn
−−
npf
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).
113
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
$.
−−
−−
−−
−−
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
HD
damit 0
−
auch 1
HD
NX 502 HD
die 2
NX 503 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 114
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: SIMPX 513 −
−
−
−
MF 511
NF 512
OA VF 507
LK 508
ON
HD
NX 500 HD
Jedes
PX 509
HD
Ja−Wort 0
OAK NX 510
−
VXFIN 501
−
OPP
zieht 1
HD
KONJ
KONJ
KONJ
NX 502
NX 503
NX 504
NX 505
NX 506
HD
HD
HD
HD
HD
Applaus 2
nach 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 515 −
−
−
−
−
MF 514 ON
OA
NX 512
NF 513
− VF 507
KONJ
LK 508
EN−ADD 509
MOD
HD
ADVX 500
VXFIN 501
ADVX 502
HD
HD
HD
Selbstverständlich
VC 510
−
hat 0
ONK
NX 503
NX 504
−
nicht 1
−
Karin 2
NX 511
OV
−
Jöns 3
die
KONJ
VXINF 505 HD
4
−
NX 506
HD
Flächen 5
−
gebucht 6
, 7
sondern
8
HD
die 9
SPD 10
. 11
12
ADV
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
ON
KONJ
NX 500
VXFIN 501
VXFIN 502
HD
HD
HD
Lausbuben
−
MF 507
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
−−
−−
−−
115
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. 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
walisischen 4
HD
ADVX 503
HD
seien
1
−
ADJX 502
HD
0
ADJX 507 −
VXFIN 501
HD
die
NX 506
HD
ADJX 500
PRED
ziemlich 5
ähnlich 6
7
ART
ADJA
NN
VAFIN
ART
ADJA
ADV
ADJD
npn
npn
npn
3pks
dpn
dpn
−−
−−
PX 503 KONJ
−
KONJ PX 502 −
HD
PX 500
NX 501
−
HD
in
und 0
um 1
Berlin 2
3
APPR
KON
APPR
NE
a
−−
a
asn
PX 505 −
HD NX 504 HD
−
ADJX 503 −
−
ADJX 500 HD
vom
NX 502 HD
4.
99
APPRART
ADJA
ADJA
CARD
dsm
dsm
dsm
−−
0
15.
ADJX 501 HD
1
2
3
116
SIMPX 521 −
−
− FKOORD 520 KONJ
KONJ
MF 517 V−MOD VF 509
LK 510
ON
−
−
HD
FDP 0
in 2
Thüringen
−
HD
knapp
4.000 5
, 6
in
7
−
Sachsen
und 10
in 11
HD
ADJX 508
HD
3.300 9
−
NX 507
HD
8
NX 516 HD
ADJX 506
HD
OA
PX 515
−
NX 505
HD
4
V−MOD
NX 514
ADJX 504
HD
3
OA
PX 513 −
ADVX 503
HD
hat 1
−
NX 502
HD
KONJ MF 519
V−MOD
NX 512
VXFIN 501 HD
Die
OA
PX 511
HD
NX 500
−
MF 518
HD
Brandenburg
2.000
12
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
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
9
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
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 KONJ
KONJ
SIMPX 510 − VF 505
HD
NX 500
VXFIN 501 HD
Der
−
LK 506
ON
−
SIMPX 511
−
HD
Fall 0
MF 507
−
VF 508
MF 509
PRED
ON
PRED
ADJX 502
NX 503
ADJX 504
HD
ist 1
−
−
brisant 2
, 3
HD
die
HD
Mischung
explosiv 6
.
4
5
ART
NN
VAFIN
ADJD
$,
ART
NN
ADJD
$.
nsm
nsm
3sis
−−
−−
nsf
nsf
−−
−−
117
7
8
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
$,
KOUS
PRF
ART
NN
NE
ADJD
VVFIN
$.
−−
−−
−−
as*3
nsf
nsf
nsn
−−
3sis
−−
118
7
.
1
ADJD
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 513 −
−
−
VF 511 OA
ON
NX 506 −
LK 507
−
HD
vorigen
Sonntag 1
−
hätte 2
PX 509
−
Michael 4
HD
ADJX 503
−
Frank 3
VC 510
−
NX 502
HD
0
FOPP
−
VXFIN 501
HD
MOD
EN−ADD 508
HD
ADJX 500
Den
−
MF 512
−
Nehr 5
NX 504 HD
am 6
OV
−
liebsten 7
aus 8
dem 9
VXINF 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:
119
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 0
LK 508
gesehen 1
NX 510
− NX 503 HD
das 3
HD
−
−
−
ADVX 504
−
ist 2
−
ADJX 509
ADJX 505
HD
Album 4
HD
wesentlich 5
schlanker 6
als 7
das 8
erste 9
. 10
11
ADJD
ADJD
VVPP
VAFIN
ART
NN
ADV
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). 120
SIMPX 511 −
−
−
VF 510 V−MOD PX 509 −
HD NX 508 APP
APP EN−ADD 505
LK 506
− NX 500
NX 501
−
Unter
HD
dem 0
Motto
ON
VXFIN 502
HD
1
MF 507
HD
HD
Kino−Extrem 2
PRED
NX 503
NX 504
−
agiert 3
HD
der 4
−
Regisseur 5
HD
als 6
Filmjockey 7
8
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.
121
NX 505 −
HD NX 504 KONJ
−
KONJ
EN−ADD 502
EN−ADD 503
−
−
NX 500 −
wie
−
Pete 0
NX 501 −
Sampras
oder
1
2
−
Yewgeni 3
Kafelnikov 4
5
KOKOM
NE
NE
KON
NE
NE
−−
nsm
nsm
−−
nsm
nsm
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-T¨ ubingen tagset (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 analyzed as a VVPP in VC. In the second example, the verbal participle has an adjectival reading and is annotated as an ADJD in MF. 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
1
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
7
wurde
8
Concerning the differences between verbal and adjectival passives in English cf. Bresnan (1995).
122
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
ADVX 511 HD
−
ADJX 504 HD
VC 512 HD
HD
ADVX 505 HD
Es
ist
schade
,
daß
so
wenig
PPER
VAFIN
ADJD
$,
KOUS
ADJA
NN
ADV
ADV
ADJD
VAFIN
nsn3
3sis
−−
−−
−−
npf
npf
−−
−−
−−
3pis
0
1
7.3
2
3
4
akademische
5
Leistungen
VXFIN 506 HD
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 analyze 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: SIMPX 509 −
−
−
VF 507
MF 508
−
ON
VC 504
V−MOD
LK 505
OV
HD
VXINF 500
VXFIN 501
HD
HD
Geplant
−
HD
NX 502 −
war 0
PX 506
HD
der 1
NX 503 HD
Papst−Besuch 2
seit 3
langem 4
. 5
6
VVPP
VAFIN
ART
NN
APPR
NN
$.
−−
3sit
nsm
nsm
d
dsn
−−
123
SIMPX 517 −
−
−
VF 515
MF 516
−
ON
V−MOD
FKONJ 513
PX 514
−
−
−
HD
MF 511
NX 512
FOPP
PRED
HD
PX 507
VC 508
−
HD NX 500
ADJX 501
HD
Auf
Pioneer 0
OV
HD
VXINF 502
VXFIN 503
HD
HD
HD
aufmerksam 1
geworden 2
−
LK 509
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 assigment 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. 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
−−
2 The identifier “HEADLINE” is automatically inserted into the comment line above the sentence for each syntactic unit which is marked as a headline in the original data.
124
SIMPX 504 −
−
MF 502
VC 503
OA
HD
NX 500
VXINF 501
HD
HD
WBM−Chefs
ablösen 0
1
NN
VVINF
apm
−−
If there is no verb at all within a headline, it is annotated like an isolated phrase (cf. 3.4.5): NX 504 HD
−
ADJX 503 −
NX
HD
ADJX
501
502
HD
HD
Handelsorganisation
vollkommen 0
NN
kopflos
1
2
ADJD
ADJD
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): SIMPX 505 −
−
VF 503
LK 504
ON NX 500
NX 501
HD
HD
Rechtschreibreform
: 0
7.5
HD 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.
125
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 −−
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
NE
np*
np*
126
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: PX 503 −
HD NX 502 APP
APP
NX 500 −
mit
HD
den 0
DM 501 −
Worten 1
" 2
−
aha
3
, 4
−
aha
5
, 6
aha
7
8
APPR
ART
NN
$(
ITJ
$,
ITJ
$,
ITJ
d
dpn
dpn
−−
−−
−−
−−
−−
−−
127
EN−ADD 503 − NX 502 HD
−
NX 500
DM 501
HD
−
Welt
−
Oh 0
, 1
no
2
3
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
Vielzahl 0
der 1
V−MOD −
VXFIN 502
HD
PX 503
HD
betroffenen 2
Mieter 3
,
HD
so
Bertermann
,
VXINF 507
HD
HD
bereits
ADJA
NN
VAFIN
PROP
$,
ADV
NE
$,
ADV
VVPP
$.
nsf
nsf
gpm
gpm
gpm
3sks
−−
−−
−−
nsm
−−
−−
−−
−−
Ein
−
−
VF 509
VF 510
LK 511
ON
HD
MOD
VXFIN 502
ADVX 503
HD
HD
NCX 501 HD
Kuratorium 0
, 1
das
2
ist 3
−
−
ON
HD
12
SIMPX 516
SIMPX 515
NCX 500
11
.
8
ART
−
10
ausgezogen
7
NN
−
9
ADVX 506
ART
−
6
OV
NX 505
HD
daher 5
VC 512
MOD HD
ADVX 504
HD
sei 4
MF 511
NX 510
HD
ADJX 501 HD
Eine
LK 509
MF 512
OA
VXFIN 505 HD
der 5
MF 514
HD
NCX 504 −
wohl 4
LK 513 PRED
Gedanke 6
NCX 506
HD
, 7
HD
macht
8
PRED
ADVX 507
ADJX 508
HD
sich 9
MOD
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
−−
−−
−−
3
On the T¨ uBa-D/Z web page (http://www.sfs.uni-tuebingen.de/en/de_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. Appendix: The T¨ uBa-D/Z Data Formats.
128
13
SIMPX 513 −
− SIMPX 512
MF 511 ON
−
−
−
VF 506 PRED
LK 507 HD
EN−ADD 508 −
LK 509 HD
ADJX 500 HD
VXFIN 501 HD
NCX 502
VXFIN 503 HD
−
−
Schön
"
,
sagte
,
"
ist
"
.
ADJD
$(
$,
VVFIN
NE
NE
$,
$(
VAFIN
PDS
PTKNEG
$(
$.
−−
−−
−−
−−
3sit
nsm
nsm
−−
−−
3sis
nsn
−−
−−
−−
7.7
2
3
4
5
6
7
8
9
das
ADVX 505 HD
"
1
Scholl
MOD
NCX 504 HD
$(
0
Mehmet
MF 510 ON
10
nicht
11
12
13
Elliptical weil and wenn auch Constructions
Both conjunctions may introduce sentential phrases as well as adjectival phrases. The sequence wenn auch apparently is in a process of grammaticalization. According to Duden (7th edition) both wenn auch as well as the subordinate conjunction weil can function as coordinating conjunctions between adjectival phrases. However we decided to follow the STTS guidelines. Consequently, we analyze weil as a subordinating conjunction (KOUS) in all cases. Furthermore the sequence wenn auch is analyzed like a conjunction which introduces an elliptical clause. This is in accordance with the distributional differences between these conjunctions and the coordinating conjunctions within elliptical constructions, cf. Karl ist ins Freibad gegangen und Max ins Hallenbad. vs. * Karl ist ins Freibad gegangen weil Max ins Hallenbad. NX 507 −
−
HD
ADJX 506 HD
− SIMPX 505 −
− MF 504 PRED ADJX 503 −
ADJX 500 HD
C 501 −
HD
ADJX 502 HD
der
herzkranke
,
weil
notorisch
ART
ADJA
$,
KOUS
ADJD
ADJA
NE
nsm
nsm
−−
−−
−−
nsm
nsm
0
1
2
3
4
eifersüchtige
5
Gary
6
129
SIMPX 514 −
−
−
− NF 513 MOD SIMPX 512 −
−
MF 510 PRED VF 506 ON
LK 507 HD
NX 500 HD
VXFIN 501 HD
ADVX 502 HD
MF 511 MOD
PRED
ADJX 508 −
ADJX 509 HD
− C 503 −
ADJX 505 HD
auch
relativ
HD
es
ist
ziemlich
,
wenn
PPER
VAFIN
ADV
ADJD
$,
KOUS
ADV
ADJD
ADJD
$.
nsn3
3sis
−−
−−
−−
−−
−−
−−
−−
−−
0
1
2
perfekt
ADVX 504 HD
3
4
5
6
7
130
langweilig
8
.
9
Chapter 8 Criteria for the Distinction of Grammatical Functions 8.1
Subcategorization of Verbs
The T¨ uBa-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). 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: 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
HD
−
VXFIN 502
ADVX 503
HD
HD
sind 4
PRED
NX 508 −
ADJX 505
HD
etwa 5
HD
ADJX 504
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
−−
−−
1
In case of interest, please refer to web page (http://www.sfs.uni-tuebingen.de/en/de_tuebadz. shtml) for contact information.
131
SIMPX 511 −
−
−
VF 509
MF 510
ON
OA
NX 505 −
LK 506
−
HD
älteren
Brüder 1
−
sich 3
−
NX 503
HD
fühlen 2
HD
NX 502
HD
0
ADJX 508
−
VXFIN 501
HD
PRED
PX 507
HD
ADJX 500
Die
OPP
für 4
HD
ihre 5
HD
ADVX 504 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
−−
−−
−−
8.1.1
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 T¨ uBa-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]./Sp¨ ater w¨ urden wir [FOPP u ¨ber 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.
8.1.2
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] F¨ uhrungen statt. [MOD-MOD Wann] finden [V-MOD am Samstag] F¨ uhrungen statt? 132
2. [MOD da] finden [V-MOD am Samstag] F¨ uhrungen statt. [V-MOD wann] finden [MOD da] F¨ uhrungen statt? [MOD dann] finden [V-MOD am Samstag] F¨ uhrungen 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/X-MOD, 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).
8.1.3
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. 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 Wortsch¨ opfungen] sind [PRED es] nur. [PRED es] sind nur [ON neue Wortsch¨ opfungen]. oder sind [PRED es] nur [ON neue Wortsch¨ opfungen]. [PRED das] sind ohnehin [ON die schw¨ achsten Partner]. [ON die schw¨ achsten Partner] sind [PRED das] ohnehin. oder sind [PRED das] ohnehin [ON die schw¨ achsten Partner]. 133
Subject-verb agreement suggests that neue Wortsch¨ opfungen und die schw¨ achsten 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 u ¨berall]. [PRED u ¨berall] 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 w¨ utender Mop]. ein w¨ utender Mop is considered the subject, because it is a noun phrase. Therefore, the prepositional phrase is PRED. 4. [ON [ON [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]. Hauptauftraggeber] ist [PRED die Bremer Verwaltung]. 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. 4.2.9).
134
References Bech, G. 1955–57. Studien u ¨ber das deutsche Verbum infinitum. Kopenhagen. 2 B¨ ande. 2. unver¨anderte auflage 1983 mit einem Vorwort von Catharine Fabricius-Hansen. T¨ ubingen: 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, pages 49-57, Sydney, Australia, 49–57. Brants, T. 1997. The NeGra Export Format for Annotated Corpora. University of Saarbr¨ ucken, Germany. Brants, T. 1998. TnT–A Statistical Part-of-Speech Tagger. Universit¨at des Saarlandes, Computational Linguistics, Saarbr¨ ucken, Germany. Bresnan, J. 1995. Lexicality and Argument Structure. In Invited Paper given at the Paris Syntax and Semantics Conference. October 12-14, 1995. URL: http://wwwcsli.stanford.edu/∼bresnan/download.html. Drach, E. 1937. Grundgedanken der Deutschen Satzlehre. Frankfurt/M. Drosdowski, G. (Ed.). 1995. Duden ”Die Grammatik der deutschen Gegenwartssprache”. Mannheim, Leipzig, Wien, Z¨ urich: 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. Grundz¨ uge der deutschen Syntax nach ihrer geschichtlichen Entwicklung dargestellt. Stuttgart. Erste Abteilung. Grewendorf, G. 1991. Aspekte der deutschen Syntax, Band 33 of Studien zur deutschen Grammatik. T¨ ubingen: Gunter Narr Verlag. Helbig, G., and J. Buscha. 1998. Deutsche Grammatik. Ein Handbuch f¨ ur den Ausl¨ anderunterricht. Leipzig, 18. Auflage. ¨ Herling, S. H. A. 1821. Uber die Topik der deutschen Sprache. In Abhandlungen des frankfurterischen Gelehrtenvereins f¨ ur deutsche Sprache, 296–362, 394. Frankfurt/M. Drittes St¨ uck. Hinrichs, E. W., J. Bartels, Y. Kawata, V. Kordoni, and H. Telljohann. 2000. The T¨ ubingen treebanks for spoken German, English, and Japanese. In W. Wahlster (Ed.), Verbmobil: Foundations of Speech-to-Speech Translation. Berlin: Springer. 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/. 135
H¨ohle, T. N. 1986. Der Begriff ‘Mittelfeld’. Anmerkungen u ¨ber die Theorie der topologischen Felder. In A. Sch¨one (Ed.), Kontroversen alte und neue. Akten des 7. Internationalen Germanistenkongresses G¨ ottingen, 329–340. Kathol, A. 1995. Linearization-Based German Syntax. PhD thesis, Ohio State University. Kiss, T. 1995. Infinitive Komplementation. Neue Studien zum deutschen Verbum infinitum. T¨ ubingen: Max Niemeyer. K¨ ubler, 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. Marcus, 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. M¨oller. 2007. Manual for the Annotation of in-document Referential Relations. University of T¨ ubingen, May 2007. Plaehn, O. 1998. Annotate - Bedienungsanleitung, Universit¨at des Saarlandes, FR 8.7 Computerlinguistik, Projekt C3 Nebenl¨aufige Grammatische Verarbeitung, Sonderforschungsbereich 378, Ressourcenadaptive Kognitive Prozesse, 13. April 1998. ¨ P¨ utz, H. 1986. Uber die Syntax der Pronominalform ’es’ im modernen Deutsch. T¨ ubingen: Stauffenburg. 2nd edition. Schiller, A., S. Teufel, and C. Thielen. 1995. Guidelines f¨ ur das Tagging deutscher Textcorpora mit STTS. Technical report, Universit¨aten Stuttgart und T¨ ubingen. 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. Technical report, Verbmobil-Report 239. Telljohann, H., E. W. Hinrichs, S. K¨ ubler, and H. Zinsmeister. 2006. Stylebook for the T¨ ubingen Treebank of Written German (T¨ uBa-D/Z). University of T¨ ubingen, July 2006. Trushkina, J. 2004. Morpho-syntactic annotation and dependency parsing of German. PhD thesis, University of T¨ ubingen. (http://w210.ub.unituebingen.de/dbt/volltexte/2004/1523).
136
Appendix: The T¨ uBa-D/Z Data Formats The T¨ uBa-D/Z treebank is released in three different data formats : 1. the NEGRA export format, 2. the Penn Treebank format, 3. an XML format (including anaphora and coreference relations). 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). An example of the NEGRA export format is given below, combined with the graphical representation of the syntactic annotation for the sentence ”Vikare mssen sich nach dem Kandidatengetz so verhalten, wie es von einem k¨ unftigen Pfarrer erwartet werden kann”. Graphical representation (print out of the annotate tool): −
SIMPX 523 −
−
−
− NF 522 PRED−MOD SIMPX 521 −
−
−
MF 520 ON
FOPP
MF 518 OA VF 512 ON
LK 513 HD
NX 500 HD
VXFIN 501 HD
Vikare
0
müssen
PX 519
V−MOD
PRED
PX 514 −
1
sich
2
NX 503 −
nach
3
dem
NN
VMFIN
PRF
APPR
ART
npm
3pis
ap*3
d
dsn
HD
4
HD
VC 515 OV
HD
NX 502 HD
−
Kandidatengetz
ADVX 504 HD
VXINF 505 HD
NX 516 −
− C 506 −
NX 507 HD
HD
ADJX 508 HD
verhalten
,
wie
es
von
VVINF
$,
KOUS
PPER
APPR
ART
ADJA
NN
VVPP
VAINF
VMFIN
$.
dsn
−−
−−
−−
−−
nsn3
d
dsm
dsm
dsm
−−
−−
3sis
−−
8
9
10
11
12
13
14
erwartet
15
werden
VXFIN 511 HD
so
7
Pfarrer
HD
VXINF 510 HD
ADV
6
künftigen
VXINF 509 HD
NN
5
einem
VC 517 OV refvc 509
OV
16
kann
Kandidatengesetz
The first line of the sentence representation (marked as ’begin of sentence’ (BOS) includes the sentence id (here: 24429), the identity of the last annotator (here the one with id 27), 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 (1222 points to article 155 of the edition of 11/7/1992). Secondary edges (here: ’refvc’ pointing from node # 510 to node # 509, a dependency within the verbal complex) as well as corrections of misspellings (here: ’Kandidatengesetz’) are also represented.
137
17
.
18
Export format: #BOS 24429 27 1134150923 1222 Vikare NN npm HD 500 mssen VMFIN 3pis HD 501 sich PRF ap*3 HD 502 nach APPR d - 514 dem ART dsn - 503 Kandidatengetz NN dsn HD 503 %% Kandidatengesetz so ADV -- HD 504 verhalten VVINF -- HD 505 , $, -- -- 0 wie KOUS -- - 506 es PPER nsn3 HD 507 von APPR d - 519 einem ART dsm - 516 knftigen ADJA dsm HD 508 Pfarrer NN dsm HD 516 erwartet VVPP -- HD 509 werden VAINF -- HD 510 kann VMFIN 3sis HD 511 . $. -- -- 0 #500 NX -- ON 512 #501 VXFIN -- HD 513 #502 NX -- OA 518 #503 NX -- HD 514 #504 ADVX -- PRED 518 #505 VXINF -- OV 515 #506 C -- - 521 #507 NX -- ON 520 #508 ADJX -- - 516 #509 VXINF -- OV 517 #510 VXINF -- OV 517 refvc 509 #511 VXFIN -- HD 517 #512 VF -- - 523 #513 LK -- - 523 #514 PX -- V-MOD 518 #515 VC -- - 523 #516 NX -- HD 519 #517 VC -- - 521 #518 MF -- - 523 #519 PX -- FOPP 520 #520 MF -- - 521 #521 SIMPX -- PRED-MOD 522 #522 NF -- - 523 #523 SIMPX -- -- 0 #EOS 24429
138
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). Graphical representation: SIMPX 517 −
−
VF 514 OA
HD
LK 511
HD
ON
HD
VXFIN 501 HD
etwas 0
, 1
ON
MF 510
HD
So
−
LK 509
ADVX 500
NX 502 −
sagen
2
, 5
NX 504 HD
hätten
6
MOD
−
HD
Abgeordneten 4
MOD
ADVX 512
VXFIN 503 HD
die 3
−
MF 516
−
NX 508 −
−
SIMPX 515
sie 7
VC 513 HD
OV
ADVX 505
ADVX 506
VXINF 507
HD
HD
HD
auch 8
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
−−
−−
−−
−−
−−
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 explicitely marking non-attached nodes. Here, the SIMPX node dominating the parenthesis (node #515) is marked as not having a mother node. #BOS 7219 19 1121695339 839 So ADV -- HD 500 etwas PIS *** HD 508 , $, -- -- 0 sagen VVFIN 3pis HD 501 die ART np* - 502 Abgeordneten NN np* HD 502 , $, -- -- 0 htten VAFIN 3pkt HD 503 sie PPER np*3 HD 504 auch ADV -- HD 505 noch ADV -- HD 506 nicht PTKNEG -- HD 512 erlebt VVPP -- HD 507 . $. -- -- 0 #500 ADVX -- - 508 #501 VXFIN -- HD 509 #502 NX -- ON 510 #503 VXFIN -- HD 511 #504 NX -- ON 516 #505 ADVX -- MOD 516 #506 ADVX -- - 512 #507 VXINF -- OV 513 #508 NX -- OA 514 #509 LK -- - 515 #510 MF -- - 515 #511 LK -- - 517 #512 ADVX -- MOD 516 #513 VC -- - 517 #514 VF -- - 517 #515 SIMPX -- -- 0 #516 MF -- - 517 #517 SIMPX -- -- 0 #EOS 7219
139
2. The Penn Treebank Format This format is based on the format of the Penn Treebank (Marcus et al. 1993). The attachment of constituents is shown via bracketing and indentation. Thus, all constituents which show the same level of indentation are attached on the same level. In the Penn Treebank format, grammatical functions, which are shown in the NEGRA export format in the column ”edge label”, are attached to the syntactic label via a colon. Thus, the label ”NX:OA” means that the constituent is a noun phrase with the grammatical function accusative object. The Penn Treebank format is a representation that combines the linear representation of words with their attachment to higher constituents. For this reason, this format is restricted to completely context-free tree structures, i.e. it cannot adequately represent the annotation of parentheses in TBa-D/Z. In order to capture the original syntactic annotation as well as the original word order in the sentence, it was decided to introduce a new edge label to mark such cases: PAR. Thus, the sentence ”So etwas , sagen die Abgeordneten , htten sie auch noch nicht erlebt .”, as shown above is represented in the Penn Treebank format by the following bracketed structure: %% sent. no. 7219 ( (SIMPX (VF (NX:OA (ADVX (ADV:HD So) ) (PIS:HD etwas) ) ) ($, ,) (SIMPX:PAR %% here starts the parenthesis! (LK (VXFIN:HD (VVFIN:HD sagen) ) ) (MF (NX:ON (ART die) (NN:HD Abgeordneten) ) ) ) ($, ,) (LK (VXFIN:HD (VAFIN:HD htten) ) ) (MF (NX:ON (PPER:HD sie) ) (ADVX:MOD
140
(ADV:HD auch) ) (ADVX:MOD (ADVX (ADV:HD noch) ) (PTKNEG:HD nicht) ) ) (VC (VXINF:OV (VVPP:HD erlebt) ) ) ) ($. .) )
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.
141
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 necessarly 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. %% sent. no. 33 ( (R-SIMPX (C (NX:ON (PRELS:HD die) ) ) (MF (NX:OA (NX:HD (ART die) (EN-ADD:HD (NN 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 prfte) ) ) ) ($. .) )
142
3. The XML Format The XML format is a custom-made XML format that follows the NEGRA export file format. It is designed to accomodate all original information provided in the export format, including e.g. comments and editor/origin information, which are resolved so that it is not necessary to consult the NEGRA tables. Dominance relations between nodes are represented directly within the XML tree structure. Root nodes are marked by the attribute parent=”0”. Thus, it is possible to represent parentheses without the use of additional labels. The root node of a parenthesis contains the attribute parent=”0”, which signifies that while the tree is part of the higher constituent where linear order is concerned, it is not attached to the surrounding tree. 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 erkl¨arte, sie werde als Kriegsgegnerin kandidieren”. The personal pronoun “sie” is anaphoric to the antecedent noun phrase “Schillen”. In the XML document, an tag is added below each node that is part of a referential relation. The tag, which is a child of the tag, encodes the type of referential relation and the node ID of the antecedent node. In our example, the antecedent is the node with ID s 1723 n 500, that is the NX dominating the named entity “Schillen”. This NX in turn is in a coreferential relationship with node s 1721 n 500 (node number 500 in sentence 1721), thus part of a coreference chain. Graphical representation of the tree without annotation of the referential relation: SIMPX 514 −
−
− NF 513 OS
−
−
VF 506 ON
LK 507 HD
VF 508 ON
LK 509 HD
NX 500 HD
VXFIN 501 HD
NX 502 HD
VXFIN 503 HD
Schillen
VC 511 OV
NX 504 −
,
sie
$,
PPER
VAFIN
KOKOM
NN
VVINF
$.
nsf
3sit
−−
nsf3
3sks
−−
nsf
−−
−−
2
3
4
als
VXINF 505 HD
HD
VVFIN
1
werde
−
MF 510 PRED
NE
0
erklärte
SIMPX 512 −
5
Kriegsgegnerin
143
6
kandidieren
7
.
8
XML format including the referential relation:
144
Index accusative object,double, 79 AcI, 78 adverbial adjective, 20, 65, 66 adverbial phrase, 24, 70 ambiguity, 17, 29, 31, 86, 88–91, 109 apposition, 25, 38 attributive adjective, 19, 20, 30, 32, 33, 65, 67
KOORD-field, 15, 16, 24, 94, 102
C-field, 15, 24, 92, 95, 99 cardinal numbers, 20, 36, 55, 56 circumposition, 20, 64 coherency, 77, 78 comparatives, 12, 119 context-freeness, 11 coordination, 12, 14, 18, 19, 24, 34, 62, 67, 101–104, 106–113, 115, 121
named entities, 11, 19, 26, 45, 51–53 Negra export format, 137 Negra treebank, 9 node labels, 11, 18, 19, 24, 26, 42, 45, 46, 73, 91, 108, 125 nominalized adjective, 66 non-ambiguity, 29, 87 non-words, 21, 56
Dependency Grammar, 30, 61 determiner phrase, 24, 60 discourse marker, 10, 12, 23, 24, 30, 125– 127
ordinal numbers, 54
lassen, 78 levels of annotation, 17, 18 long-distance dependency, 11 long-distance dependency, 29, 96, 120 longest match principle, 17, 23, 28, 111 modal verbs, 21, 85
edge labels, 11, 16, 18, 25, 26, 29, 30, 87, 101 elliptical construction, 17, 101, 116, 117 elliptical constructions, 12, 23 Ersatzinfinitiv, 15, 24, 73, 74 expletive, 25, 57, 59, 134 flat clustering principle, 17, 31, 71, 92 foreign language material, 20, 42, 53 headline, 10, 12, 23, 73, 86, 87, 124 high attachment principle, 17, 31, 107, 121 imperative, 21, 79 incoherency, 78 infinitives with zu, 75, 76, 78 initial field, 13, 15, 24, 89 isolated phrase, 26, 27, 90, 91, 113, 125 145
paratactic construction, 24, 111, 113 parenthesis, 10, 12, 23, 128 PARORD-field, 15, 16, 24, 94, 95, 113 part-of-speech tags, 11, 18, 26, 56, 62 particle verb, 81 Penn Treebank format, 137 postmodification, 31, 69 postmodifier, 17, 19, 31, 35–37, 39, 44, 49, 51, 70, 89, 99, 100, 119 postnominal modifier, 35, 36 postposition, 20, 64 predicate, 82, 131, 134 predicate-argument structure, 10, 16 predicative adjective, 20, 65, 122 preference principle, 87, 124 premodification, 31 premodifier, 31, 33, 51, 54, 65, 66, 68, 70, 89, 109 prenominal modifier, 32, 33, 35 preposition, 20, 30, 44, 47, 61, 62, 132 proper noun, 19, 20, 26, 31, 34, 37, 41, 42, 45–50, 53
punctuation marks, 21, 23, 39, 89, 90 relative clause, 15, 24, 88, 99, 100 relative clause, event-modifying, 100 relative clause, independent, 100 resumptive construction, 14, 15, 24, 96, 133 reusability, 8, 10 secondary edge label, 11, 18, 19, 25, 26, 29, 45, 48, 49, 71, 72, 78, 99 split coordination, 25, 102, 114 superlative forms, 119 syntactic dependencies, 11 T¨ uBa-D/S treebank, 9 T¨ uBa-D/Z data formats, 9, 137 theory-neutrality, 8, 10 TIGER treebank, 9 topicalization, 12, 123 topological fields, 11–18, 29, 30, 32, 82, 86, 92, 108, 116, 117 truncated word, 21, 106 verb complex, 11, 13, 15, 24, 71–73, 75, 77–79, 81, 99, 123 verb particle, 21, 81 VERBMOBIL treebank, 8, 9, 26 XML format, 137
146