Manual for the Annotation of in-document Referential Relations

Manual for the Annotation of in-document Referential Relations
Manual for the Annotation of in-document Referential Relations
Author: Karin Naumann (M.A.),
Seminar für Sprachwissenschaft, Abt. Computerlinguistik Universität Tübingen
Date: July 2006 / May 2007
Abstract
This paper presents relevant information concerning our annotation of in-document coreference
and of anaphora/cataphora. It provides a definition of the textual and semantic relation types
and of the category system used for the annotation together with a description of potential markables in the framework of coreference and anaphora/cataphora resolution. It also describes the
data base containing the annotated texts and gives an illustration of the annotation tools used for
our task. The overall aim is to provide comprehensive and comprehensible guidelines for both
users of our released data and researchers designing a similar task. Therefore, it does not only
describe the annotation background and process but also unfolds the process of discussing and
deciding on controversial cases in order to arrive at a reliable annotation standard.1
Contents
1 Introduction
1.1 Relation Types
1.2 Annotation Data
1.3 Potential Markables
1.4 Category System
2
2
2
3
4
2 Discussion of MATE Categories
6
3 Annotation of Referential Relations
3.1 Identification and Extraction of Markables
3.2 Annotation Tool
3.3 Annotation Categories
3.3.1 Coreferential
3.3.2 Anaphoric
3.3.3 Cataphoric
3.3.4 Bound
3.3.5 Split Antecedent
3.3.6 Instance
3.3.7 Expletive
3.3.8 Inherently Reflexive
10
10
10
11
11
16
20
21
28
31
35
37
4 Conclusion
38
5 References
38
1
Many thanks to Heike Zinsmeister, Holger Wunsch, Yannick Versley and Piklu Gupta for helping me refining
content and form of the text.
1
1 Introduction
1.1 Relation Types
The linguistic relations that are annotated in the texts can be subsumed under the notion of
‘referential relations’. We define referential relations as a cover-term for all contextually
dependent reference relations. The annotation is restricted to anaphoric, cataphoric and
coreferential expressions referring to a nominal or pronominal antecedent preceding or
following the dependent expression within the same text. We follow Mitkov (2003) in defining
anaphora as a linguistic phenomenon of pointing back to a textual entity preceding the anaphor,
i.e. the referring phrase. In the case of cataphora, the antecedent which is referred to is
following the cataphor. Thus, the linguistic framework underlying this definition is on the level
of text rather than on the level of semantics. Coreference, on the other hand, relates to semanticpragmatic reference of two or more nominal phrases to the same extra-linguistic referent. It
follows from this definition that there might be cases of anaphora where the referring terms are
not coreferential. These cases are called ‘identity-of-sense anaphora’, not ‘identity-of-reference
anaphora’.
In our framework, we accomplish the task of coreference resolution, thus identifying all
coreferential chains in a text. We also annotate all instances of identity-of-reference anaphora.
With respect to identity-of-sense anaphora, we annotate all instances of so-called ‘bound’
anaphora. But note that in the current release, these cases are not included in the data in order to
get a maximum of annotation consistency, this category being rather problematic. Nevertheless,
those instances have been annotated and will be published in a future release of our data.
We do not annotate other purely semantic relations, e.g. ‘part-whole’ or holonymy-metonymy.
We also disregard the annotation of event anaphora, i.e. those relations which hold between an
anaphoric expression and an underlying proposition or subsumed event rather than an
identifiable antecedent in the text. Concerning the word class, we annotate instances of
pronominal anaphora and coreferential lexical NP anaphora but neither verb and adverb
anaphora nor zero anaphora in elliptic constructions.
The inventory of the relations is inspired by the annotation scheme first developed in the MATE
project (cf. Davies et al.: 1998). Nonetheless, it only adopts those referential relations from
MATE which correspond to the definition of relation types described above.
1.2 Annotation Data
The annotation is based on the Tübingen treebank of written German (TüBa-D/Z). This
treebank uses as its data source a collection of articles of the German daily newspaper ‘taz’ (i.e.
‘die tageszeitung’). From this database, currently 1,700 data files, i.e. newspaper articles, with
36,000 sentences comprising 640,000 tokens are annotated for referential relations.
Due to its fine grained syntactic annotation, the TüBa-D/Z treebank data are ideally suited as a
basis for the identification of markables and for extracting relevant syntactic and semantic
information for each markable. For further information, please refer to the corresponding
annotation manual (Telljohann et al.: 2006).
2
1.3 Potential Markables
The markables that are subject to annotation are those linguistic elements that might refer to a
contextual antecedent. Poesio (2004) defines markables as “the text constituents that realize
semantic objects that may enter in anaphoric relations”. Hirschman and Chinchor (1997) state
“... that just because an element is ‘markable’, it does not follow that there are later references
to it – that is, it may or may not participate in coreference. [...]. The relation is marked only
between pairs of elements both of which are markables. This means that some markables that
look anaphoric will not be coded, including pronouns, demonstratives, and definite NPs whose
antecedent is a clause rather than a markable.“
Markables that will be annotated for anaphora/cataphora and for coreference are definite noun
phrases, personal pronouns, relative, reflexive, and reciprocal pronouns, demonstrative,
indefinite and possessive pronouns. All markables are extracted automatically from the TüBaD/Z treebank. With regard to the text type and text position, we annotate all parts of the
newspaper article, i.e. we annotate the proper article text as well as all headlines, subtitles and
lists, enumerations, etc. preceding/following the text or being inserted in the article.2
We annotate coreferential definite descriptions, i.e. definite NPs, including complex (e.g.
coordinated) noun phrases. The boundaries of a nominal markable are defined by the maximal
extension of this NP which includes complements as well as adjuncts of the head noun, cf.
example 1.1.
Example 1.1: TüBa-D/Z, file 235
Eine niederländische Sozialhilfeempfängerin hat in einem alten Buch, das
die Frau für umgerechnet zwei Mark auf einem Flohmarkt in Utrecht erstanden
hatte, [1 zwei Rembrandt-Radierungen] gefunden. Zu Hause fielen ihr [2 die
beiden Graphikblätter [[des Meisters] [aus dem 17. Jahrhundert]]] entgegen.
Eng.3: A welfare recipient from the Netherlands found [1 two Rembrandt
etchings] in an old book that the woman had bought at a flea market in
Utrecht for two Deutschmark in local currency. At home, [2 both graphics of
the 17th century master] fell into her hands.
The maximal NP also includes appositions and other statements in brackets, between commas
and dashes. As mentioned before, we also establish relations between elements of the headline
and the text itself. Example 1.2 and 1.3 illustrate this fact.
Example 1.2: TüBa-D/Z, file 104
[1 Stahmer]: 3.000 Schüler aus Bonn erwartet. […] Wie [2 Schulsenatorin
Ingrid Stahmer (SPD)] sagte, seien bislang jedoch „nur einige hundert“
angemeldet worden.
Eng.: [1 Stahmer]: 3.000 pupils from
senator Ingrid Stahmer (SPD)] said,
registered up to now.
Bonn expected. […] As [2 school
“only a few hundred” had been
2
Graphs, tables and illustrations are not included in the database.
The English translation of all examples in this document aims at being able to understand the sense of the
German equivalent but also at reflecting the German sentence structures in order to be able to follow the
argumentation.
3
3
An NP markable might itself be embedded in another NP. In example 1.3, markable 1
“Radunskis” functions as prenominal modifier within the complex NP “Radunskis neue
Entschlußfreude”.
Example 1.3: TüBa-D/Z, file 240
[[1 Radunskis] neue Entschlußfreude]. Kurz vor Ende der Wahlperiode dreht
[2 der Senator] noch einmal voll auf.
Eng.: [1 Radunkski’s] new delight in deciding. Shortly before the end of
the legislative period [2 the senator] gives his all.
MATE uses categories for possessive relations within a noun phrase. Example 1.4 illustrates the
relation type ‘attribute’. In example 1.5, there exists a ‘part’-relation between markable 1 and 2.
In contrast to the MATE annotation scheme (cf.: Poesio 2000), we do not annotate relations
within a phrase. You will find a brief discussion on the differences between MATE and our
annotation in section 2.
Example 1.4:
[[1 The height] of [2 the wings]]
Example 1.5:
[[1 The seat] of [2 the chair]]
Regarding the annotation of referential relations, we do not annotate indefinite nominal
descriptions containing an indefinite article or a quantifier. They may be antecedents, though.
Thus, in example 1.6 we annotate the coreference relation between [1] and [4] but not the one
between [2] and [3].
Example 1.6: TüBa-D/Z, file 303
[1 Ein Bundeswehrsoldat] ist am Dienstag in Griechenland von [2 einem NatoGegner] verletzt worden. Medienberichten zufolge schleuderte [3 ein 29
Jahre alter Grieche] in Salonica seinen Helm auf die Windschutzscheibe
eines Militärfahrzeugs, das [4 der Soldat] fuhr.
Eng.: On Tuesday, [1 a soldier of the German Federal Armed Forces] was
injured by [2 an opponent of NATO] in Greece. According to media reports,
[3 a 29-year-old Greek] in Salonica hurled his helmet onto the windscreen
of a military vehicle that [4 the soldier] was driving.
1.4 Category System
As mentioned above, the categories are based on the annotation scheme developed within the
MATE project. It is, however, an adaption of this system, i.e. those categories in MATE not
corresponding to the definition of markables and relation type described in section 1.1 and 1.2
are either avoided or redefined.
We use the following MATE categories for complex relations:
Our labels
Corresponding MATE labels
bound
split_antecedent
split_antecedent
bound anaphors
element
evoked entities
4
instance
instantiation
We do not use the following MATE categories: ‘function-value’, ‘subset’, ‘attribute’, ‘part’,
‘strict possession’, ‘event relation’, ‘situation’ and ‘empty strings’. An explanation for the use
or rejection of MATE categories can be found in section 2.
In conclusion, our category system currently comprises the following categories for the
annotation of referential relations:
•
•
•
•
•
•
•
‘coreferential’
‘anaphoric’
‘cataphoric’
‘bound’
‘split_antecedent’
‘instance’
‘expletive’
For an extensive description of the categories please refer to section 3.3.
5
2 Discussion of the MATE Category System
Since the annotation of coreferential, anaphoric and cataphoric relations apparently causes few
problems, the first step in developing an annotation standard was to discuss the MATE
categories for the annotation of ‘complex relations’.
There are two questions involved in this task:
1.) Which MATE categories should be used according to our definition of referential
relations and potential markables?
2.) Which labels should be used for identifying our own annotation categories?
In the following, the MATE categories for complex relations are defined and illustrated by
examples from the MATE annotation manual (Poesio: 2000). The illustration is accompanied
by a short explanation for our decision to use or to reject the respective category.
Bound Anaphors
The anaphor is bound by the same quantifier as its antecedent.
Example 2.1:
[1 Nobody] likes to lose [2 his] job
Example 2.2:
[1 Every man] for [2 himself]
The category is used under the same label ‘bound’ because it agrees with the definition of anaphora used in our project and differs from the description of the category ‘anaphoric’ because it
does not include a definite description as its antecedent and thus does not imply coreference.
Function-Value
Relationship between a function and its value(s).
Example 2.3:
[1 The temperature]
degrees].
rose
to
[2
90
degrees]
before
dropping
to
[3
70
Describing an exclusively semantic relation, this category is not used in our project.
Set Relations
a) Element: one discourse entity is an element of the set denoted by the other discourse entity.
Example 2.4:
[1 The kids] went to a party last weekend. [2 Paul] wanted to wear his
new suit, but [3 Jane] insisted on wearing her jeans.
The category complies with our definition of anaphora and it is used with the label ‘split
antecedent’ (formerly ‘part_of’). For further discussion of this category see section 3.3.5.
b) Subset: one discourse entity denotes a subset of the set denoted by the other discourse
entity.
6
Example 2.5:
Alors donc vous avez ici les modèles de [1 fusées]. […] Et vous allez de
vous mettre d’accord sur un classement/hein classer [2 les fusées qui
ont bien volé] ou [3 les fusées qui ont moins bien volé].
Eng.: Here you have the models of the [1 rockets]. Please classify [2
the rockets which flew well], and [3 the rockets which didn’t fly well].
The category denotes a semantic relation. Therefore, it is not used in our project.
c) Possessive Relations
Attribute: one discourse entity expresses something which is an attribute of another
discourse entity.
Example 2.6:
[[1 The height] of [2 the wings]]
On the one hand, this category denotes a semantic relation which does not correspond to our
definition of referential relation. On the other hand, it does not correspond to the definition
of markable boundaries because it is assigned to a relation within the nominal phrase.
Part: One discourse entity denotes a physical part of another discourse entity.
Example 2.7:
[[1 The seat] of [2 the chair]]
The category denotes a semantic relation within a phrase. Therefore, it is not used in our
project.
Strict possession: the relationship between two objects where one ‘belongs’ to the other.
Example 2.8:
It was a brave decision by [1 Jerry Seinfeld] to turn down $5m an
episode to make another series of [2 [3 his] hugely popular sitcom].
It is not used in our project being a semantic relation. Note that the anaphoric relation
between the possessive pronoun “his” (markable 3) and “Jerry Seinfeld” (markable 1) is
annotated.
Instantiation
The second discourse entity refers to a particular instantiation of the class identified by the first
discourse entity.
Example 2.9:
Speaker A: “We need [1 oranges].” – Speaker B: “There are [2 some] at
Corning”
This category corresponds to the definition of the relation type ‘identity-of-sense anaphora’ and
is used under the label ‘instance’ in the same sense.
7
Event Relation
Link between a discourse entity and a preceding event, expressed by a verbal phrase or
sentence.
Example 2.10:
[1 Muslims from all over the world were taught gun-making and guerrilla
warfare in Afghanistan]. [2 The instructors] were members of some of the
most radical Islamic militant groups in the region.
This category does not denote a contextual relation but it refers to event anaphora resp. frame
semantics. It is not used in our project.
Situation
Link between an item or situation and typical components of that scenario.
Example 2.11:
[1 The victorious French team] is parading [2 the World Cup] through [3
Paris] from the top of [4 a double-decker bus] as [5 a nation] continues
with [6 mass celebrations].
This category also belongs to the field of frame semantics. It does not denote a referential
relation and is not used in our project.
Propositions, Events and Actions
Link between a discourse entity and a previous proposition, event or action.
Example 2.12:
[1 The 23-year-old had hit his head against another player] during a game
of Aussie-rules football. McGlinn remembered nothing of [2 the collision],
but developed a headache and had several seizures.
Denoting event anaphora rather than referential relations on text level, this category is not used
in our project.
Evoked entities
These elements do not contain strings from the text or dialogue, but an informal description of
the evoked entity.
Example 2.13:
[1 John] arrived at 7, but [2 Mary] was much later. [3 They] missed the
film and went to the bar instead.
Being very close to the description of the set relation ‘element’ (see above), we decided to
annotate both cases under the label of ‘split_antecedent’ (formerly ‘part_of’).
Empty Strings and Clitics
For omitted elements that have to be reconstructed from the context.
8
Example 2.14:
Add [1 the dry yeast] to the water and let [2 empty] sit for a few minutes.
Example 2.15:
Speaker A: Dov'è [1 Gianni?] – Speaker B: [2 empty] È andato a mangiare.
The category is used for elliptic constructions. The relation described here is not a contextual
relation but it includes bridging assumptions and presuppositions. Thus, it is not annotated in
our project.
9
3 Annotation of Referential Relations
This section deals with the identification and extraction of markables. It gives a description of
the annotation tool(s) used for the mark-up process and explains the annotation categories
mentioned above.
3.1 Identification and Extraction of Markables
Annotation of referential relations involves two main tasks:
1.) the identification of markables, i.e. identifying the class of expressions that can enter into
referential relations
2.) the identification of the particular referential relations that two or more expressions enter
into.
Identification of markables requires at least partial syntactic annotation of the text. Markables
have to be identified semi-automatically from the output of a chunker or full parser, if
referential relations have to be annotated from plain text. If a parser is not available, the
markables have to be identified completely manually. However, in each of these two scenarios,
identification of markables is a time-consuming process. In case of semi-automatic annotation,
the effort depends on the quality of the parser, but will require at least some amount of manual
postcorrection of the parser output. Identification of markables is considerably easier for
treebank data since treebanks already provide the necessary syntactic information. For German,
there are currently two large-scale treebanks available: the NEGRA/TIGER treebank (Brants et
al., 2002) and the Tübingen treebanks for spoken and written German VERBMOBIL and TüBaD/Z (Stegmann et al.: 2000; Telljohann et al.: 2006). All the treebanks were annotated with the
help of the annotation tool “Annotate” (Plaehn, 1998). The treebank annotations are available in
three different formats: the NEGRA export format (Brants, 1997), an XML format4 (cf.
appendix of this stylebook), and the Penn treebank bracketing format5 (Marcus et al.: 1993).
3.2 Annotation Tool(s)
The annotation of referential relations is performed manually. Until the beginning of 2006, the
task was carried out by means of the annotation tool MMAX (Müller and Strube, 2003),
developed by the European Media Lab (EML)6. The abbreviation “MMAX” stands for “MultiModal Annotation in XML”. It is a stand-off annotation, i.e. base data and annotation data are
stored in separate files. The annotation categories are considered as attribute values assigned to
the respective markables.
MMAX recognizes two different relation types:
1.) Set Relation: transitive and undirected
2.) Pointer Relation: intransitive and directed
4
also see: http://www.sfs.uni-tuebingen.de/en_exportxml.shtml, visited 20.06.2006
also see: http://www.cis.upenn.edu/~treebank, visited 20.06.2006
6
cf.: http://mmax.eml-research.de, visited 20.06.2006
5
10
In our annotation, the set relation was used for the annotation categories ‘anaphoric’,
‘cataphoric’, ‘coreferential’ and ‘bound’. The pointer relation was used for the categories
‘split_antecedent’ (formerly ‘part_of’) and ‘instance’.
The annotation of referential relations with MMAX included the following steps:
•
•
•
Step 1: in the text window, the markable that is to be annotated is selected by left mouse
click (highlighted with yellow background colour). Note: you do not have to create the
markables. They are taken from the treebank and displayed in specific font colour.
Step 2: in the attribute window, an attribute value at the level ‘Type’ is chosen and assigned
to the markable
Step 3: in the text window, the relation between markable and antecedent is established by
clicking on the immediate antecedent (except for cataphora) with right mouse click and
choosing between set or pointer relation
=> Visualized result: all members of one set are highlighted (red font colour).
Since February 2006, we use the Annotation Tool ‘PALinkA’7 developed by Constantin Orasan
from Wolverhampton University8. It is a java-based application which uses XML format for the
annotation data. Beside the visualization of referential chains and different search functions, the
tool provides the possibility to add comments to each annotation which is very helpful for
revising and discussing problematic cases.
The annotation with PALinkA is carried out in the following manner:
•
Step 1: select menu ‘Tags’ from the menu bar at the top of the browser window and choose
the annotation category that you want to apply. Note: You do not have to create the markables. This information is already available from the treebank and is displayed by square
brackets plus background colour. Additionally, all markables are displayed in alphabetical
order in the right window frame.
• Step 2: click on the markable that you want to annotate. Note: if the markable is itself part
of a larger markable, you have to select the item from a menu window.
• Step 3: click on the closest antecedent (in the case of cataphora, it is the immediately
following NP)
• Step 4: add comment and save the annotation
=> Visualized result: the referential expressions receive the same background colour. When
moving the computer mouse over the anaphor, an arrow will appear pointing to the antecedent.
7
8
See: http://clg.wlv.ac.uk/projects/PALinkA/, visited 20.06.2006
See: http://clg.wlv.ac.uk/, visited 20.06.2006
11
3.3 Annotation Categories
In this section, the categories for the annotation of referential relations are defined and
illustrated by examples from the text corpus. Additionally, the section quotes problematic cases
connected with the development of an annotation standard for each annotation category.
3.3.1 Coreferential
This category is used for definite NPs in cases where two NPs refer to the same extra-linguistic
referent. This definition follows van Deemter and Kibble (2000).
Example 3.1: TüBa-D/Z, file 238
Der Vorhang geht wieder auf im [1 Metropol]. Kultursenator will [2 das
Theater] an Privatinvestor verkaufen.
Eng.: The curtain rises again in [1 the Metropol]. Culture senator wants to
sell [2 the theatre] to private investor.
Note that there are often semantic relations involved that hold between two coreferential
expressions. In this example, it is the relation of ‘instance/class’. In a forthcoming project
phase, this kind of information may be used to identify potential markables and their antecedents which could then be extracted automatically from GermaNet, a lexical-semantic net for
German (Kunze/Lemnitzer: 2002).
Discussion / Problematic cases
When defining the category system for the annotation of coreference, we decided not to
consider semantic relations between textual entities nor structural characteristics, e.g. identical
head of two or more NPs. Instead, we restricted the definition of coreference to a purely
functional, pragmatic description, i.e. reference to the same extra-linguistic referent.
The following section exhibits some problematic cases regarding coreferential relations that we
encountered during the annotation of the TüBa-D/Z treebank.
For appositions and copula constructions we made the following decisions:
Appositions belonging to the same maximal NP -> no internal reference marking
•
“Gerhard Schröder, (der) Bundeskanzler,...” -> one NP
Eng.: “Gerhard Schröder, (the) Federal Chancellor,...”
• “Gerhard Schröder, ein passionierter Zigarrenraucher,...” -> one NP9
Eng.: “Gerhard Schröder, a passionate smoker of cigars,...”
Copula Constructions:
• “Gerhard Schröder ist der deutsche Bundeskanzler.” -> definite predicate NP ->
coreferential NPs.
Eng.: “Gerhard Schröder is the German Federal Chancellor.”
• “Gerhard Schröder ist ein passionierter Zigarrenraucher.” -> indefinite predicate NP -> not
coreferential
Eng.: “Gerhard Schröder is a passionate smoker of cigars.”
9
All ‘invented’ examples that are not taken from the treebank appear without number and with normal font type.
12
Constructions with “als” (eng. “as”):
Example 3.2 illustrates that NPs with “als” (eng. “as”) do not enter into a coreferential relation.
Example 3.2: TüBa-D/Z, file10
Es gibt [1 einen neuen Kuli] auf dem Markt, der heute schon als [2 Rarität]
zu bezeichnen ist.
Eng.: There is [1 a new ball pen] on the market, which is considered [2 a
rarity] already today.
We agreed that predicative uses of NPs should generally be treated separately. Nevertheless, the
definition of a new annotation type is postponed to a further project phase.
Another problem concerns the definiteness of markables: should we also annotate indefinite
descriptions sharing a coreference relation with a textual antecedent being indefinite because of
the text type? We decided not to do so (cf. section 1.3, example 1.6).
As for generic uses of NPs, we decided not to create coreferential relations. But concerning
abstract concepts, we do annotate the nominal expression, if it appears repeatedly in the text
with the same interpretation.
Another problem concerns an NP referring to a person which acts in different roles. In example
3.3, we have to differentiate between the actor “John Travolta” (markable 1) and his role as a
“berechnender Karriereanwalt” (markable 3). In this case, we decided to put markables 1 – 5
into one set because in the given context all terms refer to the same person.
Example 3.3: TüBa-D/Z, file 180
[1 John Travolta] verklagt als Bostoner Anwalt zwei Firmen, die [2 er] für
den Leukämietod von acht Kindern verantwortlich macht. Anfangs wittert [3
der berechnende Karriereanwalt] nur die hohe Entschädigungssumme, doch ganz
allmählich wird der Fall zu einer selbstzerstörerischen Obsession.
Gerichtsdrama, Umweltthriller und großes Schauspielkino, in dem [4
Travolta] und [5 sein] Gegenspieler Robert Duvall zu Hochform auflaufen.
Eng.: [1 John Travolta] as a lawyer from Boston sues two companies which [2
he] considers responsible for the death of eight children as a result of
leukaemia. In the beginning, [3 the calculating high flying advocate] only
scents high compensation sums, but slowly the case becomes a selfdestroying obsession. Court drama, environmental thriller and great actor's
cinema, where [4 Travolta] and [5 his] antagonist Robert Duvall achieve top
form.
The next problem is related to lacking gender/number/case agreement, which is especially
important in German with its overt inflectional system. In example 3.4, markable 1 referring to
a country and markable 2 referring to its inhabitants are coreferent despite the difference in
number of the nominal head. Number, gender and case are not relevant regarding coreferential
relations as long as the terms refer to the same referent, in this example the Finnish ice hockey
team.
Example 3.4: TüBa-D/Z, file 229
Schweden und [1 Finnland] im Viertelfinale der Eishockey-WM: Das 6:1 über
die Schweiz war der zweite Sieg des Titelverteidigers, der damit genauso
4:0 Punkte hat wie [2 die Finnen] (4:1 über Weißrußland).
13
Eng.: Sweden and [1 Finland] in the quarter final of the ice hockey world
cup: the 6:1 score against Switzerland was the second victory of the
titleholder, which has 4:0 points just like [2 the Finns] (4:1 over
Belarus).
As an example for difference in case see example 3.5:
Example 3.5: TüBa-D/Z, file 202
Als Vorsitzender des HDO-Untersuchungsausschusses, der Mauscheleien von [1
SPD-Ministerpräsident Wolfgang Clement] bei der Oberhausener Trickfilmfirma
aufklären soll, hat es Meyer vom wirtschaftspolitischen Sprecher binnen
Halbjahresfrist zum Fraktionsvorsitzenden der NRW-CDU gebracht. Ohne die
Pleitenserie [2 Clements] in den letzten Monaten wäre der Karriereschub des
Duos nicht möglich gewesen.
Eng.: As chairman of the HDO commission of inquiry, which is to investigate
irregularities by [1 SPD prime minister Wolfgang Clement] at the Oberhausen
animation company, Meyer made it from the spokesman for economic policy to
the fraction leader of the NRW-CDU within a mid-year period. Without [2
Clement's] series of mishaps within the last months, the advance in the
career of the two would not have been possible.
The next case illustrates the difference between identity of two strings on text surface and
coreference between these strings. In example 3.6, there is no coreference relation between
markable 1 and 2 because of the first markable referring to the person’s name as a string of
letters, not to the person itself.
Example 3.6: TüBa-D/Z, file 236
Was der Krieg weiterhin verspricht, ist viertens Abenteuer unter Lebensgefahr. Kann man aber schneller haben, wenn man sich in Amerika eine Zigarette ansteckt. Und alles unter diesem Level gibt 's schon: Bungee-Jumping,
Wildwasser-Rafting und alle die Sachen, die hinten mit -ing aufhören wie [1
Sharping]; […]Frauen kriegen im zweitschlimmsten Fall ein Telegramm von [2
Sharping] und im schlimmsten Fall müssen sie sich die nächsten 30 Jahre
Geschichten anhören, die anfangen mit „Damals im Kosovo“.
Eng.: What the war promises is fourthly adventure at the risk of one's
life. One can achieve this faster by lighting up a cigarette in America.
And everything beneath this level is already available: bungee jumping,
wild water rafting and all those things ending in “-ing” like [1
Scharping]; […] In the second worst case, women receive a telegram from [2
Scharping] and in the worst case, within the next 30 years they have to
listen to stories beginning with “Once upon a time in Kosovo”.
In example 3.7, we consider the title of a book mentioned in the text and the book itself as being
coreferential.
Example 3.7: TüBa-D/Z, file 266
Seit er 1995 [1 die Textsammlung „Kanak Sprak“] veröffentlicht und damit
deutsch-türkische Lebensentwürfe jenseits von multikultureller Idylle und
Kulturkampfthesen vorgestellt hat, gilt er als Sprachrohr der dritten
Einwanderungsgeneration: Feridun Zaimoglu, 1964 im türkischen Bolu geboren,
seit gut 30 Jahren in Deutschland ansässig. […] In [2 „Kanak Sprak“]
schreiben Sie, die „Kanaks“ seien „unerreichbar“.
Eng.: Since publishing [1 the collection of texts “Kanak Sprak”] in 1995
and thus presenting German-Turkish concepts of living beyond a multicultural idyll and theses of cultural conflicts, he is considered the
spokesman of the third generation of immigrants: Feridun Zaimoglu, born in
14
the Turkish city of Bolu in 1964, living in Germany for almost 30 years.
[…] In [2 “Kanak Sprak”] you write that the “Kanaks” are “unreachable”.
In example 3.8, the term “Opa” (eng. grandpa) does not function as a term for family relations
but as a proper name thus entering into referential relation with coreferential markable 2 and
anaphoric pronoun (markable 3).
Example 3.8: TüBa-D/Z, file 236
So war [1 Opa] körperlich als auch mental gut auf den 14-18er vorbereitet.
Auf den 39-45er war [2 Opa] als Ortsgruppenleiter ebenfalls bestens
vorbereitet, durfte aber nicht mit, weil [3 er] schon zu alt war.
Eng.: In this way, [1 grandpa] was physically and mentally well prepared
for the 14-18 (World War I). For the 39-45 (Word War II), [2 grandpa] as
leader of a local group was likewise best prepared but was not allowed to
join it because [3 he] was already too old.
The next problem concerns nominal coordination. In example 3.9, there exist coreferential
relations between the coordinate NPs (markable 1 and 2) as well as between parts of the NP
(markable 3 and 4).
Example 3.9: TüBa-D/Z, file 103
[1 Grüne und ÖTV] kritisieren Privatisierungen [2 [3 Die Grünen] und die
ÖTV] haben die Privatisierungspolitik des CDU/SPD-Senats scharf kritisiert.
Diese habe zu einer Verunsicherung der Beschäftigten geführt und eine Blockade für notwendige Modernisierungsmaßnahmen ausgelöst, hieß es gestern in
einer gemeinsamen Presseerklärung der ÖTV-Chefin Susanne Stumpenhusen und
der Fraktionsvorsitzenden von [4 Bündnis 90/Die Grünen], Michaele Schreyer.
Eng.: [1 Green Party and ÖTV] criticize privatisation. [2 [3 The Greens]
and the ÖTV] strongly criticized the politics of privatization by the
CDU/SPD senate. It had led to uncertainty of the employees and caused a
blockade against necessary means of modernisation, it was said yesterday in
a joint press release of the ÖTV leader Susanne Stumpenhusen and the
fraction leader of [4 Bündnis 90/the Greens], Michaele Schreyer.
In example 3.10, there is no coreference relation between markable 2 and 3 because 2 functions
as a label noun embedded in the definite expression “Staatsanwaltschaft Freiburg” and is not a
referential expression.
Example 3.10: TüBa-D/Z, file 70
Aufgrund einer Anzeige des Bund für Umwelt und Naturschutz Deutschlands
(BUND) ermittelt nun [1 die Staatsanwaltschaft [2 Freiburg]]. Der Verdacht,
daß es sich um nicht zugelassene Maissorten handelt, wurde von der
Chemischen Landesuntersuchungsanstalt in [3 Freiburg] bestätigt.
Eng.: Because of a complaint by the German Federation for Environment and
Nature Conservancy (BUND) [1 the Public Prosecutor's Department [2
Freiburg]] is investigating now. The suspicion that it concerns prohibited
sorts of maize has been confirmed by the federal centre for chemical
investigation in [3 Freiburg].
In example 3.11, markable 2 does not stand in a coreferential relation to markable 1 since they
are both different and refer to different types of substances.
Example 3.11: TüBa-D/Z, file 1236
[1 Dieses Ozon] wirkt genauso wie [2 das Ozon weiter oben] als Filter für
die UV-Strahlen.
15
Eng.: [1 This ozone] like [2 the other ozone further above] works as a
filter for ultraviolet rays.
In example 3.12, markable 1 and 2 are coreferential.
Example 3.12: TüBa-D/Z, file 1291
Die Berliner Staatsanwaltschaft hat [1 einen Polen und zwei Österreicher]
wegen unerlaubten Umgangs mit Kernbrennstoffen verhaftet. [2 Die drei
Beschuldigten] sollen fünf Kilogramm Cäsium 137 […] eingeführt haben,
erklärte die Justiz.
Eng.:The Berlin public prosecutor's office arrested [1 a Pole and two
Austrians] because of dealing illegally with nuclear fuel. [2 The three
accused] are supposed to have imported five kilogram of caesium 137 […],
said the law.
In example 3.13, markable 3 is not coreferential to markable 2 because in this sentence the
quality of being an “Ossi” (East German) refers to the past. This past is over and therefore the
pronoun “sie”, which refers to the present, does not stand in any relation to markable 3, which
belongs to the past. Markable 2 is anaphoric, markable 3 is not coreferential.
Example 3.13: TüBa-D/Z, file 1078
Denn [1 sie] gehen ja nicht glücklich in der neuen Westgesellschaft auf und
vergessen, daß [2 sie] [3 Ossis] waren.
Eng.:Because [1 they] are not totally wrapped up in the new western society
and forget that [2 they] used to be [3 East Germans].
16
3.3.2 Anaphoric
This attribute value is used for definite pronouns referring back to a contextual antecedent. A
set relation is established between the pronoun and its antecedent. In example 3.14 the NP is
referred to by two reflexives (markable 2 and 5) and two personal pronouns (markable 3 and 4).
Thus, markables 1 – 5 belong to the same set.
Example 3.14: TüBa-D/Z, file 336
[1 Ein klarer Ton] breitet [2 sich] aus, warm und satt, bis [3 er] den
ganzen Saal erfüllt. Dann dünnt [4 er] aus, zerbröselt und verflüchtigt [5
sich].
Eng.: [1 A clear sound] spreads [2 itself] out, warm and full, until [3 it]
fills the whole hall. After that, [4 it] thins out, falls into pieces and
fades [5 itself].
Discussion / Problematic cases
There was a discussion regarding the annotation of reflexive pronouns because in German there
are many cases of ‘inherently reflexive verbs’, as e.g. “sich ereignen” (eng. “happen (itself)”).
In these cases, the reflexive is not annotated because it does not refer. There are various tests for
deciding whether a verb is inherently reflexive, e.g. coordination tests (“Sie wäscht sich und
ihre Schwester”, eng.: “she washes herself and her sister” -> not inherently reflexive). These
tests work out all right for clear-cut cases. But there are lots of boundary cases where this
decision can not easily be made. Therefore, we decided to insert a comment for all reflexives
that are not annotated for belonging to an inherently reflexive verb. Thus, it is possible to check
all cases in a revision phase after the annotation process. Additionally, the annotator refers to a
list of verbs included in the IMSLEX lexicon (cf. Eckle-Kohler, 1999 and Fitschen, 2004) for
deciding on these cases. See also chapter 3.3.8 on inherently reflexive pronouns.
The next problem concerns personal pronouns. First person plural pronoun “wir” (eng. “we”) in
direct speech cannot easily be related to a speaker because it may include individuals which are
not mentioned in the text. It is ambiguous, who the speaker includes in this group of people.
Therefore, in example 3.15 we only establish an anaphoric relation between markable 3 and 4.
But note that uses of ‘majestic plural’ are excluded from this rule, because they clearly refer to
the speaker of the direct speech thus receiving the value ‘anaphoric’ or ‘cataphoric’.
Example 3.15: TüBa-D/Z, file 149
[1 Izet] deutet auf die Plastiksandalen, die [2 ihm] geblieben sind. „[3
Wir] haben nur noch das, was [4 wir] auf dem Leibe tragen.“
Eng.: [1 Izet] points to the plastic sandals, that have stayed with [2
him]. “[3 We] only have the things, that [4 we] are wearing on our bodies.”
The next case is closely related to the last example. In example 3.16, there exists an anaphoric
relation between markable 1 and 3, not 1 and 2, although the NP “meine Wenigkeit” stands for
“me”. Remember: we establish anaphoric relations between elements of the text surface (in this
example first person personal pronoun and first person possessive pronoun) not semantic
relations.
Example 3.16: TüBa-D/Z, file 266
[1 Ich] sehe das eher als Plattform, die es möglich macht, daß Imran und [2
[3 meine] Wenigkeit] im Stakkato literarische Texte raushauen, Musikdarbietung inbegriffen.
17
Eng.: [1 I] consider this rather a platform that makes it possible that
Imran and [2 [3 my] humble self] churn out literary texts in staccato,
including musical performances.
The
next topic concerns items closely related to personal and reflexive pronouns. In the TüBaD/Z treebank, the intensifier “selbst” (eng. “self”) is analysed as an adverb and consequently
does not belong to the group of potential markables. Therefore, it does not enter into an
anaphoric relation. In example 3.17, only markable 3 is annotated referring anaphorically to
markable 1.
Example 3.17: TüBa-D/Z, file 120
Sollen [1 sie] mich doch bloß als Narren ansehen. Werden sehen, ob [2 [3
sie] selbst] nicht welche sind.
Eng.: Let [1 them] consider me a fool. (They) will see, if [2 they [3 themselves]] aren't ones.
The next problem is related to the field of fixed expressions. The following two examples 3.18
and 3.19 exemplify idiomatic uses of pronouns. In these cases, the possessive pronouns are not
annotated for anaphora since they do not refer to other lexical items within the text. Also
compare inherently reflexive verbs (see above) and fixed expressions as e.g. “sich fragen”, “mit
sich bringen”, “sich zeigen”, etc.
Example 3.18: TüBa-D/Z, file 033
„[Mein] Gott, müssen
gebackener Bräutigam.
wir viele
Kinder kriegen“,
sorgt sich
ein frisch
Eng.: “[My] God, how many children we must have”, the newly wed groom
remarks sorrowfully.
Example 3.19, TüBa-D/Z, file 251
Und so pinseln und kleben, knipsen, schweißen
Künstler, daß es nur so [seine] Art hat...
und
knobeln
jetzt
die
Eng.: And so the artists brush and paste, cut off, weld and toss, that it
only has [it's] way10...
With regard to interrogative pronouns, we only annotate those cases where these elements
function as a relative pronoun, as e.g. in: “Ich interessiere mich für [1 das], [2 was] du sagst.“ –
eng.: “I am interested in [1 that] [2 which] you are saying.” In this case, markable 2 stands in a
‘bound’ relation to markable 1.
In example 3.20, markable 2 is anaphoric.
Example 3.20, TüBa-D/Z, file 1224
Bis es mit [1 den Steinen] beschwert wird, mit [2 denen] dem Mann das
Gesicht zertrümmert wurde.
Eng.: Until it gets weighted down with [1 the stones] [2 which] were used
to smash the man's face.
As example 3.21 illustrates, lacking agreement (difference in number and gender for markable 1
and 2) does not impede the use of the category ‘anaphoric’.
10
The German idiomatic expression ‘seine Art haben’ means ‘it is a delight’
18
Example 3.21, TüBa-D/Z, file 1224
[1 Der ANC] will nicht verhandeln, [2 die] wollen uns nur umbringen.
Eng.: [1 The ANC] does not want to negotiate, [2 they] just want to kill
us.
In example 3.22 markable 2 refers to markable 1 ‘anaphorically’, because “103 albanische
Flüchtlinge” is a concrete group and therefore a concrete referent. Thus it is like all numbers not
to be annotated as ‘bound’ (cf. Example 3.23 and 3.24).
Example 3.22, TüBa-D/Z, file 1231
Italien hat am Donnerstagmorgen [1 103 albanische Flüchtlinge] abgeschoben,
[2 die] am Vorabend im Hafen von Brindisi angekommen waren.
Eng.: On Thursday morning Italy deported [1 103 Albanian refugees], [2 who]
had arrived the evening before in the harbour of Brindisi.
Examples 3.23 to 3.30 provide examples, in which markable 2 is anaphoric to markable 1.
Example 3.23, TüBa-D/Z, file 1269
Das sind [1 90 Prozent des Nationalerbes an Ikonen], [2 das] sich Anfang
der 80er Jahre noch im Land befand.
Eng.: These are [1 90 percent of the national heritage of icons] [2 which]
were still in the country at the beginning of the eighties.
Example 3.24, TüBa-D/Z, file 1108
[1 Zwei Drittel der siebzig Millionen Amerikaner], [2 die] nicht als Wähler
registriert sind, haben ein unterdurchschnittliches Einkommen.
Eng.: [1 Two thirds of the seventy million Americans] [2 who] were not
registered as voters have an income below average.
Example 3.25, TüBa-D/Z, file 1148
Damit ist [1 der Sender], [2 der] für die Einspeisung des Films in das ARDNetz verantwortlich ist, verurteilt worden […]
Eng.: As a result [1 the television station] [2 which] is responsible for
feeding the film in the ARD-net has been condemned […]
Example 3.26, TüBa-D/Z, file 1236
Das ist [1 ein ganz entscheidender Effekt], auf [2 den] ich ganz deutlich
hinweisen möchte.
Eng.: This is [1 a very crucial effect] [2 which] I want to point out
clearly.
Example 3.27, TüBa-D/Z, file 1238
Deshalb werden die Komitees auch die Aufgabe haben, dafür einzutreten, daß
[1 eine besondere Körperschaft] für die neuen Bundesländer geschaffen wird,
in [2 welche] einzelne Persönlichkeiten […] gewählt werden […].
Eng.: Therefore the committees will have the task to speak out in favour of
[1 a particular corporation] for the newly-formed German federal states for
[2 which] single personalities […] are elected […].
Example 3.28, TüBa-D/Z, file 1206
19
[1 Das 75-Seiten-Papier], [2 dessen] langatmige Gründlichkeit die mitwirkende Hand deutscher Diplomatie verrät, legt die wichtigsten Neuerungen
dar.
Eng.: [1 The 75-pages-document], [2 whose] wordy carefulness reveals the
helping hand of German diplomacy, presents the most important innovations.
Example 3.29, TüBa-D/Z, file 1253
[1 Der Kern], um [2 den] das Karussell sich drehte, löst sich somit auf.
Eng.: [1 The core] on [2 which] the roundabout rotates therefore dissolves.
Example 3.30, TüBa-D/Z, file 1262
[1 Harald Szeemann] stellt dem utopischen Deutschland [2 seine] „Visionäre
Schweiz“ entgegen.
Eng.: [1 Harald Szeeman] opposes to the utopian Germany [2 his] “Visionary
Switzerland”.
In sentences like example 3.31 one has to check carefully if the “sich” is actually ‘anaphoric’ or
‘inherently reflexive’. To see where ‘inherently reflexive’ applies see chapter 3.3.8.
Example 3.31, TüBa-D/Z, file 1352
Da haben wir ihm doch allen Ernstes vorgerechnet, daß [1 er] [2 sich] von
einer Computerfirma […] um etwa 2,8 Millionen Mark über's Ohr hat hauen
lassen.
Eng.: We listed him in all seriousness that [1 he] got [2 himself] ripped
off by a computer company […] for about 2.8 Million Deutschmark.
Although the quantifier “keine” is used in example 3.32, markable 2 stands only in an anaphoric
relation to markable 1. This is the case because the relative pronoun “die” refers to the noun
“Forderungen” and not to the quantifier.
Example 3.32, TüBa-D/Z, file 1084
Brasilien hat [1 keine der Forderungen] erfüllt, [2 die] mit den deutschen
Krediten aus dem Jahr 1988 verbunden waren.
Eng.: Brazil did not fulfil [1 any of the demands] [2 which] were connected
to the German credits of 1988.
20
3.3.3 Cataphoric
A cataphoric relation holds between a pronoun referring to a following antecedent within the
same or the following sentence. In contrast to the category ‘anaphoric’, the syntactic hierarchy
overrules contextual criteria. Consequently, the attribute value is assigned to a definite pronoun
preceding the antecedent of the main clause even if the extra-linguistic referent has already been
mentioned within the preceding text.
In example 3.33, the personal pronouns “mich” (markable 1) and “er” (markable 3) receive the
attribute value ‘cataphoric’, in the first case referring to the proper name “Ulrich Görlitz”
(markable 2) and in the second case to the personal pronoun “er” (markable 5). On the other
hand, markable 4 and markable 5 are annotated as ‘anaphoric’ both referring to markable 2.
Example 3.33: TüBa-D/Z, file 120
„Sollen sie [1 mich] als Narren ansehen!“ [2 Ulrich Görlitz] diente
freiwillig in Hitlers Armee. […] Aber [3 er] habe [4 sich] eben 1944 zur
Armee gemeldet, um der Einberufung durch die Waffen-SS zu entgehen, erzählt
[5 er].
Eng.: “Let them consider [1 me] a fool!” [2 Urlich Görlitz] served in
Hitler's army voluntarily. […] But [3 he] had enlisted [4 himself] in the
army in 1944 in order to escape the call up into the Waffen-SS, [5 he]
explains.
Discussion / Problematic cases
Generally, the concept of cataphora can be defined in two different ways:
there exits a cataphoric relation between a referential expression and its contextual antecedent,
if
1.) the antecedent has not been mentioned in the text yet (strict definition based on discourse
level)
2.) the antecedent follows the referential expression within the sentence (weak definition based
on sentence level)
We decided in favour of definition 2 in order to get the maximum of referential information. In
a further working phase, it is still possible to sort out those cases that do not conform to
definition 1.
In example 3.34, markable 2 receives the attribute value ‘cataphoric’ referring to the following
antecedent markable 3 although the NP has already been mentioned in the heading of the article
(markable 1). Relations between elements of the heading and elements of the article
transgressing the boundaries of textual parts are predominated by internal relations within one
textual part.
Example 3.34: TüBa-D/Z, file 102
Nach 222 Tagen Mahnwache am Alexanderplatz zogen [1 die Traktorenwerker aus
Schönebeck] gestern ab. Ein Investor aus Nordrhein-Westfalen will 190
Arbeitsplätze sichern
222 Tage währte [2 ihre] kleine Traktoren-Mahnwache am Alexanderplatz. Mit
drei landwirtschaftlichen Nutzfahrzeugen und ständiger personeller Präsenz
im nebenstehenden Campingmobil dauerprotestierten [3 die Beschäftigten der
Landtechnik Schönebeck (LTS)] sowie MitarbeiterInnen des Tochterunternehmens GS Fahrzeug- und Systemtechnik für die Übernahme ihres Betriebes
durch einen West-Investor.
21
Eng.: After 222 days of solemn vigil at Alexanderplatz, [1 the workers on
tractors from Schönebeck] withdrew yesterday. An investor from “NordrheinWestfalen” wants to save 190 jobs. [2 Their] little solemn vigil of
tractors at “Alexanderplatz” lasted 222 days. With three agricultural
commercial vehicles and a permanent personal presence in the neighbouring
camper [3 the employees of the “Landtechnik Schönebeck (LTS)”] as well as
employees from the subsidiary “GS Fahrzeug- und Systemtechnik” were protesting permanently for the takeover of their company by a Western investor.
Note that further anaphoric expressions without a following nominal antecedent within the same
sentence receive the attribute ‘anaphoric’ as in example 3.35 where markable 3 refers back to
markable 2.
Example 3.35: TüBa-D/Z, file 111
Daß [1 sie] von dieser Klausel Gebrauch machen würden, daran ließen [2 die
eingeflogenen Damen und Herren] von Anfang an keinen Zweifel. Nur mit Mühe
ertrugen [3 sie] den Versuch eines studentischen Personalratsvertreters,
einen Mißbilligungsantrag gegen den FU-Kanzler vorzubringen.
Eng.: That [1 they] would use this clause, [2 the ladies and gentleman
flown in] left no doubt from the beginning on. Only with effort did [3
they] suffer the attempt of a student representative of the personnel board
to introduce a vote of no confidence against the chancellor of the FU.
In example 3.36, markable 2 is cataphoric.
Example 3.36: TüBa-D/Z, file 1295
Als „Open Air Festivalorchester“ lädt das in [1 seiner] Existenz gefährdete
[2 Rundfunkorchester Berlin] am 18. Juli zu einem Strauß-Konzert vor dem
Französischen Dom ein.
Eng.: As “Open Air Festival Orchestra” the in [1 his] existence endangered
[2 Broadcasting Orchestra Berlin] asks to a Strauß-concert in front of the
French Dome on the 18th of July.
22
3.3.4 Bound
This category is used for set relations holding between a definite pronoun and a quantified or
indefinite noun phrase/pronoun as its antecedent. The anaphor “is bound by the same quantifier
as its antecedent” (Poesio 2000) (e.g. ‘some’ or ‘many’, etc.). As mentioned in section 1.1, this
category is applied to all cases described here but is not included in the current release.
In example 3.37, this relation holds between markable 2 and the indefinite pronoun (markable
1).
Example 3.37: TüBa-D/Z, file 124
[1 Wer] einen Sitzplatz haben will, [2 der] muß um 19 Uhr, wer im Treppenhaus noch etwas hören will, sollte spätestens um 20 Uhr da sein.
Eng.: [1 Whoever] wants to get a seat, [2 he] has to be there at 7 pm, whoever wants to hear something from the stairway should be there at 8 pm at
the latest.
In case that there are further pronouns following the indefinite antecedent, these markables
receive the attribute ‘anaphoric’, as shown in example 3.38. Markable 2 refers to the indefinite
NP “Die meisten Benutzer” (markable 1) receiving the value ‘bound’, whereas markable 3
receives the value ‘anaphoric’ referring back to markable 2.
Example 3.38: TüBa-D/Z, file 94
[1 Die meisten Benutzer] kaufen [2 sich] [3 ihre] Tassen selbst.
Eng.: [1 Most of the users] buy [2 their] cups [3 themselves].
Discussion / Problematic Cases
One question concerning this category was how to treat generic uses of NPs (e.g. “der Deutsche
an sich, der…”; “ein Mensch, der…”). We decided to restrict the use of the category ‘bound’ to
quantified non-referential antecedents but to include the impersonal 3rd person singular
pronoun “man”.
It is important to differentiate between indefinite NPs including a quantifier which refer to an
unspecified class of objects/persons and indefinite NPs referring to a specific group.
In the first example, markable 2 gets the attribute value ‘bound’ referring to an unspecific
indefinite group whereas in the second example it receives the value ‘anaphoric’ referring to an
indefinite but specific group of people.
•
•
[1 Einige Frauen] qualifizieren [2 sich] über spezielle Frauenförderungsmaßnahmen weiter.
– Eng.: [1 Some women qualify [2 themselves] via special support schemes for women.
[1 Einige Frauen] kamen auf mich zu, [2 die] mir unzählige Fragen stellten.
– Eng.: [1 Some women] came to me [2 who] asked innumerable questions.
Not all NPs including quantifiers are non-referential. Example 3.39 includes a quantified NP
which refers to a specific group of people. Markable 2 therefore receives the value ‘anaphoric’,
not ‘bound’.
23
Example 3.39, TüBa-D/Z, file 489
Der Selfmade-Diplomat kehrte mit [1 den drei in Jugoslawien inhaftierten
Soldaten] aus Belgrad zurück. […] [2 Sie] waren Mitglieder der dort
stationierten UN-Friedenstruppen.
Eng.: The self-made diplomat returned from Belgrade with [1 the three soldiers arrested in Yugoslavia]. […] [2 They] were members of the UN peace
corps based there.
As example 3.40 illustrates, lacking agreement (difference in number for markable 1 and 2)
does not impede the use of the category ‘bound’ (cf. example 3.4).
Example 3.40, TüBa-D/Z, file 172
„Wir werden doch [1 kein altes Hafenbecken] zuschütten“, sagte er, [2 die]
seien doch der Reiz der geplanten neuen Hafen-City.
Eng.: “We will fill up [1 no ancient port basin]”, he said, [2 they] are
the attraction of the planned new port city.
For our future work, we will create a list of text indicators including all possible surface triggers
for this category.
Example 3.41, TüBa-D/Z, file 1127
[1 Man] sei nicht gegen die U-Bahn, [2 man] würde sie im Gegenteil
begrüßen, wenn [3 man] einen schnellen Ausbau der Linie garantiert bekomme
[…].
Eng.: [1 We] are not against the tube, [2 we] would on the contrary welcome
it, if [3 we] could get a faster extension of the tube line […].
Since “man” is an indefinite pronoun, the different “man” are linked in the same sentence by
the label ‘bound’.
In examples 3.42 – 3.46, markable 2 is bound.
Example 3.42, TüBa-D/Z, file 1153
Mit all diesen obskuren Äußerungen fällt Frau Rogendorf nicht nur den
Lesben/Frauen auf den Wecker, die ihr leben selbstbestimmt und -bewusst
leben, sondern auch [1 all den Frauen/Lesben], [2 die] sich für die Abschaffung des § 218 einsetzen.
Eng.: With all her obscure remarks Mrs. Rogendorf does not only get on the
nerves of the lesbians/women who are self-confident, but also on those of
[1 all the women /lesbians] [2 who] make an effort to abolish § 218.
Example 3.43, TüBa-D/Z, file 1121
Denn [1 wen] er in seinen Club aufnimmt, [2 der] kriegt auch bald seine
Ringe.
Eng.: Because [1 who] he lets join his club, will soon get his rings.
Example 3.44, TüBa-D/Z, file 1276
Aus ihm kommen Frauen und Männer, die die Welt nach „richtig“ und „falsch“
einteilen und seit den letzten 20 Jahren [1 das] dazwischen schieben, „[2
was] beliebt“.
Eng.: From this (system) descend women and men, who divide the world into
“right” and “false” and who put in between “[1 whatever] pleases”.
24
Example 3.45, TüBa-D/Z, file 1099
Täten die Schriftsteller es nicht, wer spräche sonst wohl von den ganz
normalen Träumen, von den Träumen [1 derer], [2 die] nicht einmal einen
ruhigen Ort zum Schlafen haben […].
Eng.: If the authors did not do it, who else would talk about the ordinary
dreams, the dreams of [1 those], [2 who] do not even have a quiet place to
sleep.
Example 3.46, TüBa-D/Z, file 1212
Aber [1 alles], [2 was] den Charakter einer Partei trägt und
Ostdeutschland beschränkt bleibt, würde die Spaltung aufrechterhalten.
auf
Eng.: But [1 everything] [2 which] supports the character of a party and is
restricted to East Germany would maintain the division.
Although markable 2 in example 3.47 refers to a number, it is annotated as ‘bound’, since “etwa
1300 Atommüllfässer” does not represent a concrete referent (cf. example 3.48, 3.49).
Example 3.47, TüBa-D/Z, file 1215
In Gorleben lagern nämlich [1 etwa 1300 Atommüllfässer], von [2 denen]
mindestens 300 Mitte der achtziger Jahre im belgischen Mol waren […].
Eng.: [1 Approximately 1300 barrels of nuclear waste] are stored in
Gorleben, and at least 300 of [2 them] were in the Belgian Mol in the mid80s.
In examples 3.48 – 3.68, markable 2 is bound.
Example 3.48, TüBa-D/Z, file 1231
Im vergangenen Jahr hatte Italien [1 rund 20.000 albanische
flüchtlinge] zwangsweise in [2 ihre] Heimat zurückgebracht.
Eng.: Last year Italy brought by force [1
refugees] back into [2 their] home country.
about
20.000
Boots-
Albanian
boat
Example 3.49, TüBa-D/Z, file 1233
In den USA warteten [1 mehr als 2.500 Häftlinge] in der Todeszelle auf [2
ihre] Hinrichtung.
Eng.: In the USA [1 more than 2.500 prisoners] waited in the death cell for
[2 their] execution.
Example 3.50, TüBa-D/Z, file 1238
[1 Viele Menschen in den neuen Bundesländern] fühlen sich nach [2 ihrer]
Hochstimmung im Jahr 1990 als Menschen zweiter Klasse […].
Eng.: [1 Many people in the newly-formed German federal states] feel as
second-class citizens […] after being excited / in [2 their] high spirits
in 1990.
Example 3.51, TüBa-D/Z, file 1240
[1 Alle Konstruktionen], und mögen [2 sie] auch noch so ausgefeilt sein,
erweisen sich als wenig nützlich.
Eng.: [1 All constructions] and may [2 they] already be filed very much
turn out to be not very useful.
Example 3.52, TüBa-D/Z, file 1240
25
Europas Zukunft hängt vor allem von dem Willen und dem Mut [1 derjenigen]
ab, [2 die] die Mittel haben, um sie zu sichern.
Eng.: Europe's future depends above all on determination and courage of [1
those] [2 who] have the means to secure it.
Example 3.53, TüBa-D/Z, file 1216
[1 Über 7.000 Feuerteufel] zündeln jährlich in Berlin, legen Häuser und
Wohnungen in Schutt und Asche. [2 Ihre] Motive sind, so vorhanden,
verschieden […].
Eng.: [1 More than 7.000 fire bugs] play with fire in Berlin ever year,
raze houses and flats to ground. [2 Their] motives are different, if they
exist at all […].
Example 3.54, TüBa-D/Z, file 1246
Insgesamt starben in den USA im Jahr 1985 nicht [1 weniger als 408.000
Menschen] an den Folgen [2 ihrer] Nikotinabhängigkeit.
Eng.: All in all, not [1 less than 408.000 people] died of the consequences
of [2 their] nicotine addiction in the USA in 1985.
Example 3.55, TüBa-D/Z, file 1253
Allein der Weg dahin ist weit und mühsam und dauert ganz offensichtlich
länger [1 als jene „wenigen Wochen“], [2 die] Camdessus […] vorsieht.
Eng.: However, the path is long and difficult and obviously takes longer [1
than those few weeks] [2 which] Camdessus […] has in mind.
Example 3.56, TüBa-D/Z, file 1206
Minderheiten im Sinne der KSZE […] sind allerdings nur [1 solche], [2 die]
die Staatsbürgerschaft eines „Mehrheitslandes“ besitzen.
Eng.: Minorities as defined by the KSZE […] are only [1 those] [2 who] have
the citizenship of a “country of majority”.
Example 3.57, TüBa-D/Z, file 1263
Beim Sommerschlußverkauf der übriggebliebenen Staats- und Parteiverlage
spielen sich Dinge ab, für die [1 jeder Poker-Spieler] von [2 seinen]
Partnern hinausgeworfen worden wäre.
Eng.: At the summer sales of the remaining publishing houses of government
and party things are going on for which [1 every poker player] would have
been thrown out by [2 his] partners.
Example 3.58, TüBa-D/Z, file 1263
[1 Kein Fraktionszimmer], wo [2 sich] nicht die […] abgedruckten Leserbriefe stapeln […].
Eng.: [1 No parliamentary party room] where […] printed reader's letters do
not pile up [2 themselves] […].
Example 3.59, TüBa-D/Z, file 1101
Und da gab es [1 manche], [2 die] sich aus Schrecken noch einmal angepaßt
haben.
Eng.: And there were [1 quite a few] [2 who] adapted themselves once more
out of fear.
Example 3.60, TüBa-D/Z, file 1101
26
Er sei zwar nicht dafür, alles zu konservieren, aber „ein bißchen von [1
dem], [2 was] gewesen ist, sollten wir erhalten“.
Eng.: He is not in favour for preserving everything, but we should keep up
a bit of [1 what] has been.
Example 3.61, TüBa-D/Z, file 1139
Der schmerzhafte Entschluß zur Flucht, der Vertrauensbruch mit [1 denen],
[2 die] blieben, hat plötzlich keinen Sinn mehr.
Eng.: The painful decision to escape, the breach of trust with [1 those] [2
who] remained suddenly does not make sense any more.
Example 3.62, TüBa-D/Z, file 1176
Wenn ich merke, [1 jemand] ist besonders aufgeregt oder geladen, begrüße
ich [2 ihn] möglichst freundlich.
Eng.: When I notice that [1 someone] is particularly nervous or mad then I
bid [2 him] welcome as friendly as possible.
Example 3.63, TüBa-D/Z, file 1185
[1 Wer] [2 sich] bei der andauernden Hitze in einem naturbelassenen See
kühlen möchte, hat in Hamburg noch immer große Auswahl.
Eng.: [1 Who] wants to refresh [2 himself] in a natural lake because of the
continual heat has still a large range in Hamburg.
Example 3.64, TüBa-D/Z, file 1191
Rudolph nagte an diesem Vorwurf sehr, war er doch in der Schwimmbranche als
Trainingsweltmeister bekannt, als [1 einer], [2 der] sich […] ins Bassin
warf um die richtige Linie zu finden.
Eng.: Rudolph had to cope with this reproach, because he was known to be
the world champion of training in the swimming business, to be [1 one] [2
who] threw himself […] into the pool to find the right way.
Example 3.65, TüBa-D/Z, file 1333
Es ist wohl so, daß [1 der], [2 der] ständig im juristischen Ausnahmezustand agiert, das Gefühl für Rechtsstaatlichkeit verliert.
Eng.: It is perfectly true that [1 a person] [2 who] constantly acts in a
legal state of emergency loses the sense for the rule of law.
Example 3.66, TüBa-D/Z, file 1164
Die jetzigen Verkehrsverhältnisse überschreiten die Grenzen [1 dessen], [2
was] die Anwohner […] an Verkehrsbeeinträchtigungen hinzunehmen haben,
deutlich.
Eng.: The present traffic situation clearly exceeds the limit of [1 that]
[2 which] the residents […] have to accept of the impeding of traffic.
Uncertain Bound-Occurrences (there is still room for discussion)
Example 3.67, TüBa-D/Z, file 1206
[1 Ungefähr ein Fünftel der Menschen], [2 die] gegenwärtig in den Industrieländern leben, werden möglicherweise durch den Tabak sterben […].
Eng.: [1 Approximately one fifth of the people] [2 who] live at the moment
in the industrial nations will possibly die of tobacco […].
27
Example 3.68, TüBa-D/Z, file 1406
Bei [1 allem Verständnis], [2 das] er für die Unterzeichner des Aufrufes
habe, bekennt er ehrlich: „Eine Ostpartei ist unsinnig.
Eng.: With [1 all the understanding] [2 that] he has for the signatories of
the appeal he honestly confesses: “A party of the East is silly.
In example 3.69, markable 3 is cataphoric to markable 2 and markable 3 is bound by 1. The
numbering in example 3.69 only matches the German sentence since the English sentence
construction is different.
Example 3.69, TüBa-D/Z, file 1046
Allein in Kroatien leben inzwischen [1 an die 300.000 bosnische Kriegsflüchtlinge]. [2 Ihres] Lebens können [3 sie] einigermaßen sicher sein.
Eng.: In the meantime [1 approximately 300,000 Bosnian refugees of war]
already live in Croatia. [2 They] can be fairly sure of [3 their] lives.
In example 3.70 – sd, markable 2 is bound.
Example 3.70, TüBa-D/Z, file 1390
Da verflüchtigt sich ein Staat, und es gibt [1 welche], [2 die] wollen das
nicht wahrhaben.
Eng.: A state disappears and there are [1 some] [2 who] refuse to accept
it.
Example 3.71, TüBa-D/Z, file 1116
Aufgrund des gleichen Paragraphen gibt es in Warschau inzwischen [1
Tausende kommunaler Sozialmieter], [2 die] ihre Zahlungen eingestellt haben
[…].
Eng.: On account of the same paragraph there are meanwhile [1 thousands of
local socially disadvantaged tenants] in Warsaw [2 who] stopped their
payments […].
Example 3.72, TüBa-D/Z, file 1154
[1 Welcher Jugendliche] ist denn heute noch bereit, [2 sich] selber langfristig zum Anhängsel zu machen?
Eng.: [1 Which young person] is nowadays still prepared to make [2 himself]
a long-term appendage?
28
3.3.5 Split_antecedent (formerly ‘part_of’)
The split_antecedent relation holds between coordinate NPs (e.g. “Jane and Mary”) or plural
pronouns (e.g. “both”) and pronouns/definite NPs referring to one member of the plural
expression. The category is not applied to semantic relations of the type ‘possessive relations’.
In example 3.73, the indefinite pronoun “Sie” (markable 3) refers to both, the NP “eine junge
Frau” (markable 1) and the coordinate NP “ihre gebrechliche Mutter und vier Kinder”
(markable 2). A split_antecedent relation is established pointing from markable 3 to both of the
antecedents.
Example 3.73: TüBa-D/Z, file 634
„Vor allem die letzten Stunden waren fürchterlich“, sagt [1 eine junge
Frau], die [2 ihre gebrechliche Mutter und vier Kinder] über die Grenze
führt. [3 Sie] sind zu Fuß gekommen, denn das Auto wurde [4 ihnen] von
serbischen Freischärlern abgenommen.
Eng.: “Above all the last hours were terrible”, said [1 a young woman] who
is leading [2 her invalid mother and four children] across the border. [3
They] have come on foot because the car was taken from [4 them] by Serbian
irregulars.
Note that there is an anaphoric relation between markable 4 “ihnen” and markable 3 “Sie”.
Discussion / Problematic cases
In the MATE project, possessive relations in the sense of semantic relations are annotated (see
section 2). We decided not to annotate this kind of relation but to introduce the category
‘split_antecedent’ for reference to plural pronouns in the sense of the MATE category ‘evoked
entities’ and the set relation category ‘element’.
In the beginning of the project phase, the category was used under the label ‘part-of’ (cf.
Hinrichs et al.: 2004, 2005). The antecedents received the attribute value ‘part_of’ and a pointer
relation to the plural NP was established. A further step after discussing this category again was
to reverse the relation, i.e. the plural expression receives the value ‘split_antecedent’ pointing to
the two or more antecedents.
In example 3.74, markable 3 receives the attribute value ‘split_antecedent’ and points to both
markable 1 and 2.
Example 3.74, TüBa-D/Z, file
taz: [1 Sabrije], hast du noch Kontakt zu [2 deiner serbischen Freundin]?
Sabrije: Seit den Nato-Angriffen haben [3 wir] nicht mehr miteinander
gesprochen.
Eng.: taz: [1 Sabrije], are you still in contact with [2 your Serbian
friend]? Sabrije: Since the NATO attacks, [3 we] haven't spoken to each
other.
In example 3.75, markable 3 is not ‘part-of’ markable 1, but enters into a coreferential relation
with markable 2.
Example 3.75, TüBa-D/Z, file 217
Dort lebt asketisch ein alter Zen-Meister mit [1 seinen beiden Schülern Kibong, einem jungen Mönch, und [2 Haejin, einem Kind]]. […] manchmal
regnet es; oder der Milchzahn [3 des kleinen Jungen] wackelt.
29
Eng.: An old Zen master lives there ascetically with [1 his two disciples Kibong, a young monk, and [2 Haejin, a child]]. […] sometimes it rains; or
the milk tooth of [3 the young boy] wobbles.
In example 3.76, markable 3 has split antecedents and therefore points to markable 1 and 2. If
there was a coordinate NP [der All Jazz Club und das Villa], we could establish an anaphoric
relation between markable 3 and the complex NP.
Example 3.76, TüBa-D/Z, file 216
Nach Medienberichten müssen Ende Mai wegen neuer Eigentümer auch [1 der All
Jazz Club] und Ende Juni [2 das Villa] schließen. [3 Beide] sind seit den
fünfziger Jahren für ihre prominenten Gast-Konzerte bekannt.
Eng.: According to the news, at the end of May [1 the All Jazz Club] and at
the end of June [1 the Villa] have to close because of new owners. Since
the fifties, [3 both] have been well-known for their prominent concerts.
In example 3.77, markable 3 is a split_antecedent.
Example 3.77, TüBa-D/Z, file 1176
In Hamburg sind [1 68700 Menschen] auf soziale Fürsorge angewiesen. Hinzu
kommen [2 21300 Aussiedler und Asylbewerber]. Die Hansestadt unterstützt [3
sie] mit jährlich 650 Millionen Mark.
Eng.: [1 68,700 people] are dependent on social care in Hamburg. [2 21,300
emigrants and asylum seekers] are added. The Hanseatic city assists [3
them] every year with 650 million Deutschmark.
30
3.3.6 Instance
In cases where a specific pronoun or NP refers to a particular instantiation of the class identified
by an NP, the category ‘instance’ is assigned to the preceding or following pronoun/NP
referring to that nominal antecedent. Thus, the specific expression is considered a specific
instantiation of a class of entities. In example 3.78, the pronoun “jene” (markable 2) receives
the value ‘instance’ and points to the NP “viele Banken” (markable 1).
Example 3.78: TüBa-D/Z, file 414
Doch [1 viele Banken], vor allem [2 jene] im Staatsbesitz, sitzen auf
Haufen von faulen Krediten.
Eng.: But [1 many banks], above all [2 those] belonging to the state, are
sitting on a heap of rotten credits.
Discussion / Problematic Cases
In example 3.79, there is a purely semantic relation between markable 1 and 2. Thus, markable
2 is not annotated as instance of markable 1.
Example 3.79, TüBa-D/Z, file 239
Kurz vor Ende der revolutionären Demonstration knüppelte [1 die Polizei] in
die Menge, während Steine und Glasflaschen in Richtung [2 der BeamtInnen]
geworfen wurden.
Eng.: Shortly before the end of the revolutionary manifestation, [1 the
police] were clubbing the crowd while stones and glass bottles were thrown
in the direction of [2 the officers].
The following list represents all cases annotated so far. As you will notice, there are not very
many. The list also includes those cases that were rejected after discussion of each sentence.
Nevertheless, they are included in the list in order to illuminate the process of developing our
standard. A corresponding remark will accompany these examples.
In example 3.80, markable 2 is annotated as instance of markable 1 and markable 3 is bound.
Example 3.80, TüBa-D/Z, file 046
Jungen Linienrichtern wird bei [1 zweifelhaften Entscheidungen] - also [2
solchen], [3 die] gegen BU ausfallen - schon mal angedroht, sie ins
Internat zurückzuschicken.
Eng.: For [1 dubious decisions] - namely [2 those] [3 which] go against BU
– young linesmen are threatened to be sent back to boarding school.
In example 3.81, markable 2 is annotated as instance of markable 1.
Example 3.81, TüBa-D/Z, file 055
Die russische Diplomatie ist unentbehrlich, wenn es darum geht, [1 die zu
erwartenden Störmanöver Miloevic'] zum Scheitern zu bringen. [2 Solche
Manöver] sind todsicher zu erwarten, denn die G-8-Erklärung enthält
keinerlei Details und keine Fristen, die den Rückzug der „militärischen,
polizeilichen und paramilitärischen Kräfte aus dem Kosovo“ regeln würden.
Eng.: The Russian diplomacy is indispensable when [1 the expected disruptive actions of Milosevic] have to be impeded. [2 Such actions] are dead
sure because the G8 declaration contains no details and no deadlines that
31
regulate the fallback of the “military, police and paramilitary forces from
Kosovo”.
In example 3.81, markable 2 is annotated as instance of markable 1.
Example 3.82, TüBa-D/Z, file 230
Die konservativen Kräfte warten ja nur darauf, ihm [1 Sätze] um die Ohren
zu hauen wie [2 jenen] von den 16 Mittelstrecklern, denen er in vier Wochen
die Viererkette beibringe.
Eng.: The conservative powers are just waiting to bombard him with [1 sentences] like [2 the
one] about the 16 middle-distance runners who he is teaching the double full-back formation
within four weeks.
In example 3.83, markable 2 is annotated as instance of markable 1 and markable 3 is bound. It
is not annotated, because markable 2 cannot be considered as being an instantiation of the class
“Phtalate” which is referred to by the quantified expression.
Example 3.83, TüBa-D/Z, file 691
Außerdem, so Axel Singhofen, Giftstoff-Experte von Greenpeace in Brüssel,
sollte die Bestimmung [1 nur sechs Phtalate] betreffen - hauptsächlich [2
solche], [3 die] gar nicht in Spielzeug verwendet werden.
Eng.: Besides, said Axel Singhofen, expert on toxins for Greenpeace in
Brussels, the determination was meant to concern [1 only six phthalates] mainly [2 those] [3 which] are not used for toys.
In example 3.84, markable 2 is annotated as instance of markable 1.
Example 3.84, TüBa-D/Z, file 815
Ich könnte der taz auf die Sprünge helfen, denn Böll und ich begründeten
einst das Kartell, das weder eins war noch sich in Unterschriften
erschöpfte, sondern Solidarität mit Verfolgten leistete, was [1 so mancher
Spötter], und [2 deren] gab es so viele wie Sandflöhe am Meer, jeweils dann
wußte, wenn ihm Ungemach drohte - in solchen Fällen baten die tapfren
Schreiberlein beim „Kartell“ um Hilfe.
Eng.: I could get the taz going because Böll and I once founded the cartel
that neither was one nor was it restricted to signatures; instead it showed
solidarity with the haunted which [1 many a mocker], and [2 of those] there
existed as many as jigger fleas by the sea, knew exactly when being in
trouble - in these cases the brave little writers asked the “cartel” for
help.
In example 3.85, markable 2 is annotated as instance of markable 1. It is not annotated because
the antecedent is a quantified NP, thus markable 2 is not an instantiation of the entire class.
Example 3.85, TüBa-D/Z, file 848
Es gibt [1 2.000 Termitenarten] auf der Welt, [2 die meisten] in den Tropen
oder Subtropen, selten auch in Trockenzonen.
Eng.: There are [1 2,000 species of termites] in the world, [2 most of
them] in the tropics and subtropics, rarely also in the arid environment.
In example 3.86, markable 2 is annotated as instance of markable 1. It is not annotated because
the antecedent is a quantified NP, thus markable 2 is not an instantiation of the entire class.
32
Example 3.86, TüBa-D/Z, file 852
[1 Alle siebzehn AnwältInnen Öcalans] sind Verfechter der kurdischen Sache.
[2 Viele] stammen aus dem Umfeld des Menschenrechtsvereins IHD, der durch
seine Nähe zur PKK auffiel.
Eng.: [1 All seventeen advocates of Öcalan] are defenders of the Kurdish
case. [2 Many] originate from the periphery of the human rights association
IHD which attracted the attention because it is close to the PKK.
In example 3.87 – example 3.92, markable 2 is annotated as instance of markable 1.
Example 3.87, TüBa-D/Z, file 854
„[1 Die Hosen] an“ hatte sie nur im übertragenen Sinne, im wirklichen Leben
aber trug sie niemals [2 welche].
Eng.: “[1 The trousers] on”11 she had only in a figurative sense; in real
life she never wore [2 any].
Example 3.88, TüBa-D/Z, file 929
Eine Arbeit mit [1 gelegentlichen Sternstunden]. [2 Eine] erlebte er, als
ihm eine Kundin überschwenglich siebzig Mark Trinkgeld in die Hand drückte.
Eng.: A job with [1 occasional highlights]. [2 One] he experienced when a
client exuberantly gave him seventy Deutschmark tip.
Example 3.89, TüBa-D/Z, file 1129
„Bei [1 den Schädlingsbekämpfern] gibt es solche und solche“, erklärt Dr.
Scheurer, „[2 einige] spritzen einfach los!“.
Eng.: “Among [1 the pest controllers] there are those and others”, explains
Dr. Scheurer, “[2 some of them] just spray!”
Example 3.90, TüBa-D/Z, file 1265
[1 Neun in Folie verpackte Platten], eben der vermeintliche Plastiksprengstoff, fand der Autofahrer, der äußerst vorsichtig [2 eine der Platten] aus
der Tasche zog […].
Eng.: The car driver found [1 nine plastic wrapped sheets], the supposed
plastic explosives, and he pulled [2 one of the sheets] extremely carefully
out of the bag […].
Example 3.91, TüBa-D/Z, file 1260
Zwischen 1905 und 1918 gab Blei [1 eine Reihe von meist kurzlebigen
Zeitschriften] heraus, [2 einige] davon in wertvoller bibliophiler Aufmachung […].
Eng.: Between 1905 and 1918 Blei published [1 a series of mostly shortliving magazines], [2 some of them] in a valuable fine edition […].
Example 3.92, TüBa-D/Z, file 1259
Vierhundert Fliesen - von [1 denen] [2 circa die Hälfte] unbedruckt blieb ließ Delvoye […] einlegen.
Eng.: Delvoye arranged that four hundred tiles
approximately half of [1 them]] remained blank.
11
were
inlaid
–
and
[2
Note: The German idiom ‘die Hosen anhaben’ corresponds to eng. ‘to wear the breeches’.
33
In example 3.93, markable 2 is anaphoric to markable 1 and markable 3 is annotated as instance
of markable 2.
Example 3.93, TüBa-D/Z, file 1345
Mein Freund Thomas aus Berlin entwickelte aus seiner derzeitigen Situation
„[1 Gedichte] aus dem zwölften Haus“, [2 deren] [3 eines] hier auszugsweise
zitiert sei […].
Eng.: My friend Thomas from Berlin produced “[1 Poems] from the 12th house”
out of his present situation, [3 one] of [2 those] is cited here in parts
[…].
In example 3.94 – example 3.95, markable 2 is annotated as instance of markable 1.
Example 3.94, TüBa-D/Z, file 1130
[…] es bestünden in der Bevölkerung weit verbreitete Ekelgefühle gegenüber
[1 Ratten], so daß die Haltung [2 einer solchen] gegen den Hausfrieden
verstoßen könne.
Eng.: […] among the population there are widely spread feelings of disgust
at [1 rats] so that keeping [2 one of those] could violate the domestic
peace.
Example 3.95, TüBa-D/Z, file 1065
Nun gehe es darum, auch in Deutschland [1 mögliche Kontrollregelungen] zu
treffen […]. Entsprechende Gespräche bezüglich [2 einer solchen Kontrollregelung] würden […] wieder aufgenommen werden.
Eng.: Now the thing is to make [1 possible supervision] arrangements […].
Relevant talks concerning [2 such a supervision arrangement] would be taken
up again […].
Conclusion: The anaphor is not annotated as ‘instance’ if the antecedent does not concern the
whole class of objects but a quantified portion of that class.
34
3.3.7 Expletive
This category is different from all the other annotation categories described so far. It is an
attribute value assigned to the impersonal third person singular pronoun “it” (German “es”) and
does not denote a specific type of referential relation. On the contrary, it is used for those cases
where the pronoun is non-referential. In the literature, this is also know as ‘pleonastic IT’. But
note that it is not used for event anaphora, i.e. uses of “it” where the pronoun refers to an
underlying proposition, as e.g. in “Gasoline prizes are rising again and I do not like [it]”.
In German, impersonal 3d person sing. pronoun ‘ES’ is used as:
1.) Personal Pronoun:
[1 Das Baby] liegt in der Wiege. [2 Es] schläft ruhig.
Eng.: [1 The baby] is lying in the cradle. [2 It] is sleeping calmly.
2.) Subject of weather verbs and verbs with missing agent:
[1 Es] regnet.
Eng.: [1 It] is raining.
[1 Es] trug ihn aus der Kurve.
Eng.: [1 It] threw him out of the bend.
3.) Anticipating pronoun in extraposed sentences:
[1 Es] ist gut, dass Peter kommen konnte.
Eng.: [1 It] is good that Peter could come.
4.) Expletive pronoun in sentence-initial position:
[1 Es] kamen zwei Männer zur Tür herein.
Eng.: [1 It] came two men through the door.
[1 Es] wurde bis zum Morgen getanzt.
Eng.: [1 It] was danced until morning12.
We decided to use the category for the cases 2 and 4. The category is applied to presentational
“es” as in example 3.96
Example 3.96: TüBa-D/Z, file 120
Die Deutschen sehnen sich zurück: [Es] muss mal wieder Krieg sein.
Eng.: The Germans are longing to have something back: [It] has to be war
again.
It is used for “es” in sentence-initial position, cf. example 3.97.
Example 3.97:
[Es] waren zwei Königskinder.
Eng.: [It] were two royal children.
It is also applied to cases where the “es” is a substitute for the missing subject, cf. example 3.98.
Example 3.98: TüBa-D/Z, file 500
12
The German sentence would correspond to ‘here was dancing’ in English.
35
Dennoch gelingt [es] Christian Ebert
Briefroman Bühnenleben einzuhauchen.
mit
seiner
Inszenierung,
dem
Eng.: Nevertheless, [it] is managed by Christian Ebert with his production
to give stage life to the epistolary novel.
Discussion / Problematic Cases
We compared our approach to that of Boyd et al. (2005) Evans (2000) and Paice/Husk (1987)
although idiosyncrasies are common in these cases and some of the phenomena described in
these publications are restricted to the English language. On the other hand, many of the
German examples cannot easily be transferred to English. Besides, there are also syntactic
characteristics included in the annotation of non-referential “it”. Consequently, two types of
expletive ‘ES’ are already marked in the syntactic annotation of the German treebank: ‘VorfeldES’ and ‘Korrrelat-ES’13. These cases are not annotated separately in the process of the
annotation of referential relations being inherently defined as ‘expletive’.
In the following, we list some of the cases from the TüBa-D/Z treebank that were discussed and
which exhibit the difficulties in translating expletive ‘ES’ to another language:
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
13
“Es findet sich” -> idiomatic, not annotated (eng.: it will work out)
“Es geht (mir) um...” -> expletive (eng.: it is about...)
“Es ist mir ernst” -> expletive (eng.: it is serious for me)
“Es kam dazu” -> expletive (eng.: it happened that.)
“Es gibt mehrere Lösungen.” -> expletive (eng.: there are several solutions)
“Es handelt sich um...” -> “Es” = expletive; “sich” = inherently reflexive (not annotated)
(eng.: it concerns...)
“Es geht nach seinem Willen” -> expletive (eng.: it is done according to his will)
“Es ist Krieg” -> not annotated (eng.: it is war)
“Es waren 765” -> expletive (eng.: it were 765)
“Es geht los” -> not annotated (eng.: it is starting)
“Es fällt schwer” -> expletive (eng.: it is hard)
“Es scheint als ob...” -> expletive (eng.: it seems as if...)
“Es steht schlecht um ihn” -> expletive (eng.: it is bad for him)
“Es macht nichts” -> not annotated (eng.: it doesn’t matter)
“Es fragt sich” -> “Es” = expletive; “sich” = inherently reflexive (not annotated) (eng.: it
has to be asked.)
“Es hat sich ergeben, dass..” -> “Es” = expletive; “sich” = inherently reflexive (not annotated) (eng.: it arose that...)
“Es ist das erste Mal” -> expletive (eng.: it is the first time)
“Es ist ein Mädchen!” -> cataphoric (eng.: it is a girl)
“Es ist so, wie es ist.” -> first “es” not annotated, second “es” anaphoric (eng.: it is as it is)
For a detailed description see manual on syntactic annotation of the TüBa-D/Z treebank.
36
3.3.8 Inherently Reflexive
Reflexive pronouns, which occur with inherently reflexive verbs like “sich ereignen”, are not
annotated as anaphoric in the TüBa-D/Z. In spite of the above mentioned tests, there are many
borderline cases. Therefore these reflexive verbs get a comment in the comment-field of the
annotation tool so that they can be found easily later on.
An additional help is the verb list from the IMSLex-Lexicon (Eckle-Kohler, 1999; Fitschen,
2004). The information in this list is partially generated automatically from the corpora so that
this information can help to decide difficult cases, but it is not to be regarded as finally certain.
Test Criteria
Source: Dudenredaktion (1998, §195f.)
The Duden calls them “true reflexive verbs”. With these verbs the reflexive pronoun:
1.)
cannot be omitted.
Er schämt sich. (He is ashamed.)
*Er schämt.
2.)
cannot be replaced by another pronoun or noun.
Er schämt sich.
*Er schämt ihn.
3.)
cannot be coordinated.
Er schämt sich.
*Er schämt sich und den Nachbarn.
4.)
cannot be asked for.
*Wen schämt er?
5.)
cannot be negated.
*Er schämt nicht sich, sondern den Nachbarn.
6.)
is subject to particular limitations of the position.
*Sich schämt er.
Notice:
There are verbs which are exclusively reflexive (‘Reflexive’) and those which occur only in a
single meaning as really reflexive. The Duden calls the later ones “partially-reflexive verbs”.
7.)
(7) Partially-reflexive verbs
Sich ängstigen (‘Angst empfinden’) versus jmdn. ängstigen (‘in Angst versetzen’)
Sich schicken (‘schicklich sein’) versus jmdm. etw. schicken (‘senden’)
Also: ärgern, aufhalten, entscheiden, verlassen
Reciprocal verbs and other apparent problem cases are also covered by the tests.
8.)
9.)
(8) The reflexive pronoun stands in an interrelation.
Peter und Maria lieben sich.
(9) There are also three-digit verbs (examples from Helbig and Schenkel (1991, 58f.))
37
a. Er vereinigt den Turnverein mit dem Sportverein.
b. Die KPD vereinigt sich mit der SPD.
c. Die KPD und die SPD vereinigen sich.
38
4 Conclusion
The precise definition of referential relations and the adaption of the annotation categories to
critical cases discussed in the project guarantees a consistent treatment of markables and
standardized annotation of the data corpus. Additionally, an Inter-Annotator Agreement was
carried out as check for consistency of the annotation. For this purpose, an additional annotator
marked-up 70 files out of 766. These files were chosen randomly (file 1, 11, 21…) and there
existed no advisory exchange between the annotators and other project members. The result of
the check can be seen in Versley (2006).
5 References
Boyd, Adriane, Whitney Gegg-Harrison, and Donna Byron (2005): “Identifying non-referential
it: A machine learning approach incorporating linguistically motivated patterns.” In:
Proceedings of the ACL Workshop on Feature Engineering for Machine Learning in
Natural Language Processing (pp. 40-47). Ann Arbor, Michigan: Association for
Computational Linguistics.
Brants, Sabine, Stefanie Dipper, Silvia Hansen, Wolfgang Lezius, and George Smith (2002):
“The TIGER treebank.” In: Hinrichs, Erhard and Kiril Simov (eds.): Proceedings of the
First Workshop on Treebanks and Linguistic Theories (TLT 2002) (pp. 24–41). Sozopol,
Bulgaria.
Brants, Thorsten (1997): “The NeGra Export Format for Annotated Corpora.” Universität des
Saarlandes, Computational Linguistics, Saarbrücken, Germany.
Davies, Sarah, Massimo Poesio, Florence Bruneseaux, and Laurent Romary (1998): “Annotating Coreference in Dialogues: Proposal for a Scheme for MATE.” MATE: 1998.
Dudenredaktion (ed.). (1998): “Duden. Grammatik der deutschen Gegenwartssprache”. 6.
edition, volume 4. Mannheim, Leipzig, Wien, Zürich: Dudenverlag.
van Deemter, Kees and Rodger Kibble (2000): “On coreferring: Coreference in MUC and
related annotation schemes.” In: Computational Linguistics, 26(2) (pp. 629–637). MIT
Press.
Eckle-Kohler, Judith (1999): “Linguistisches Wissen zur automatischen Lexikon-Akquisition
aus deutschen Textcorpora.” Berlin: Logos Verlag.
Evans, Richard (2001): “Applying Machine Learning Toward an Automatic Classification of
It.” In: The Journal of Literary and Linguistic Computing 16;1. (pp. 45-57). Oxford:
Oxford University Press.
Fitschen, Arne (2004): “Ein computerlinguistisches Lexikon als komplexes System.” In: AIMS
Vol. 10 (3). University of Stuttgart: Doctoral Dissertation.
Helbig, Gerhard, and Wolfgang Schenkel (1991): “Wörterbuch zur Valenz und Distribution
deutscher Verben” 8. edition. Tübingen: Max Niemeyer Verlag.
Hinrichs, Erhard, Sandra Kübler, Karin Naumann, Heike Telljohann, Julia Truschkina (2004):
“Recent Developments in Linguistic Annotations of the TüBa-D/Z Treebank.” In:
Proceedings of the Third Workshop on Treebanks and Linguistic Theories (TLT) (pp.
51-62). Tübingen.
Hinrichs, Erhard, Sandra Kübler und Karin Naumann (2005): “A Unified Representation for
Morphological, Syntactic, Semantic, and Referential Annotations.” In: Proceedings of
the ACL Workshop on Frontiers in Corpus Annotation II: Pie in the Sky (pp. 13-20).
Ann Arbor, Michigan: Association for Computational Linguists: MIT Press.
Hirschman, Lynette and Nancy Chinchor (1997): “MUC-7 Coreference Task Definition.” In:
MUC-7 Message Understanding Conference Proceedings.
39
http://www.itl.nist.gov/iaui/894.02/related_projects/muc/proceedings/muc_7_toc.html
(last visited: 24.06.2006, no print version available)
Kunze, Claudia and Lothar Lemnitzer (2002): “GermaNet – representation, visualization,
application.” In: Proceedings LREC 2002, main conference, Vol V. (pp. 1485-1491). Las
Palmas.
Marcus, Mitchell P., Beatrice Santorini, and Mary Ann Marcinkiewicz (1993): “Building a
Large Annotated Corpus of English: The Penn Treebank.” In: Computational
Linguistics, 19 (2) (pp. 313—330). MIT Press.
Mitkov, Ruslan (ed.) (2003): “The Oxford Handbook of Computational Linguistics”. Oxford:
Oxford University Press.
Müller, Christoph and Michael Strube (2003): “Multi-level annotation in MMAX.” In:
Proceedings of the 4th SIGdial Workshop on Discourse and Dialogue (pp.198-107).
Sapporo.
Paice, Chris D. and Husk, Gareth D. (1987): “Towards the Automatic Recognition of
Anaphoric Features in English Text: The Impersonal Pronoun ‘It’.” In: Computer
Speech and Language, 2 (pp.109-132). Paris: Elsevier Academic Press.
Plaehn, Oliver (1998): “Annotate Bedienungsanleitung”. Universität des Saarlandes,
Sonderforschungsbereich 378, Projekt C3, Saarbrücken, Germany, April 2003.
Poesio, Massimo (2000): “Coreference”. In: Mengel, A., Dybkjaer, L., Garrido, J.M., Heid, U.,
Klein, M., Pirrelli, V., Poesio, M., Quazza, S., Schiffrin, A., and Soria, C. 8. Januar
2000. MATE Dialogue Annotation Guidelines.
http://www.ims.uni-stuttgart.de/projekte/mate/mdag/ (last visited: 24.6.2006)
Poesio, Massimo (2004): “The MATE/GNOME Scheme for Anaphoric Annotation, Revisited.”
In: Proceedings of SIGDIAL. Boston.
http://cswww.essex.ac.uk/staff/poesio/publications/SIGDIAL04.pdf
(last visited: 24.06.2006)
Schumacher, Helmut, Jacqueline Kubczak, Renate Schmidt and Vera de Ruiter (2004):
“VALBU – Valenzwörterbuch deutscher Verben (Studien zur Deutschen Sprache).”
Tübingen: Narr.
Stegmann, Rosmary, Heike Telljohann, and Erhard W. Hinrichs (2000): “Stylebook for the
German Treebank in VERBMOBIL.” Tübingen: Technical Report 239, Verbmobil.
Telljohann, Heike, Erhard W. Hinrichs, Sandra Kübler and Heike Zinsmeister (2006):
“Stylebook for the Tübingen Treebank of Written German (TüBa-D/Z).” Seminar für
Sprachwissenschaft, Universität Tübingen, Tübingen, Germany.
Versley, Yannick (2006): “Disagreement Dissected: Vagueness as a Source of Ambiguity in
Nominal (Co-)Reference.” Workshop Paper, ESSLLI 2006. Málaga.
40
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Related manuals

Download PDF

advertisement

Languages