OWL modelling Christian Chiarcos 080410
Annotation Guidelines
PoCoS { Potsdam Coreference Scheme: Core Scheme
Draft 0.912
Christian Chiarcos & Olga Krasavina
May 4, 2005
complex-np
A description is complex if it contains more than one noun phrase
Markables including appositions, relative clauses, etc. are more complex in their nature and
therefore receive a special tag in the scheme: "complex-np"(see 3). A simple test can be used,
in order to determine if an NP is complex: if a description consists of more than one
nominal (not pronominal) phrase, consider it a complex description. Nominal phrases
"are phrases headed by nouns and able to function as complement in clause structure: "the dog
barked"(subj), "I found the dog" (obj), "this is a dog" (predicative). " (cf. Longman)
(21.a) its ¯nancial adviser - not a compex NP (its and ¯nancial are adjectives)
(21.b) of NBI shares - NBI cannot be modi¯ed, cf. the NBI shares, very popular NBI shares:
the, very popular modify shares, rather than NBI. Thus, NBI is not an independent
noun phrase.
(21.c) all [the preferred stock] - the same
Prototypical complex NPs are:
(22.a) the surplus or pro¯t required under Delware law for payment of the dividend
(22.b) NBI, a maker of word processing systems
(22.c) its ¯nancial advisor and investment banker
Border-line case:
(23) the stock's holder - a complex np (the stock's is an independent NP)
complexity
Markable Features: set of tags used in the core scheme
Below the tags employed in the implemented core annotation scheme are outlined, together with
their short descriptions. For more detailed description, consider previous sections. The tags
correspond to annotation features and their values. Values marked with * are secondary markables.
1
1
1
1
1
1
The scheme distinguishes between primary and secondary markables. Primary markables are
always subject to annotation. Secondary markables are only annotated if they happen to serve
as antecedents for primary markables.
² inde¯nite descriptions
² other markables that are not labelled as primary markables but serve as antecedents
for them (cf. ex. 20, p. 7). Note that this does not apply to sentences!
So, if the antecedent of an anaphor is an inde¯nite expression, declare it as a new markable. In this
case, inde¯nite descriptions can be markables as well. Otherwise, do not annotate inde¯nite
descriptions as markables, even if they are in some (non-anaphoric) relation to other discourse
entities.
Inde¯nite descriptions include
² nominal descriptions with an inde¯nite article (a dog)
² nominal descriptions without an article (business, two companies)
Cataphora
We distinguish two types of forward-referring expressions, discourse cataphora and syntactic cat-
aphora.
Extensions of NPForm
np-form
Pronouns
Pronouns include personal pronouns, demonstrative pronouns, pronominal adverbs, and personal
pronouns and both in nominal use, see ¯g. 1 in Appendix for detailed list of forms. In German,
demonstrative pronouns include: dieser, jener, der, die, derjenige, and the like
Note that demonstrative pronouns such, in German solch and so-called "generic pronouns" (we,
you, they, in German wir, du, sie (without speci¯c reference), man, einer) are considered as
inde¯nite, thus no primary markables.
direct-speech
text-level (default) reference on the text level, i.e. reference NOT into or within
the quoted material
dir reference into or within direct speech
indir reference into or within indirect speech
direct_speech
A description is definite if it contains a determiner both2 (2a, 2b), a demonstrative (2c) or
possessive pronoun (2d) or a genitive attribution (2e). Quanti¯ed NPs (some of them, all the
members) are annotated as either de¯nite or inde¯nite, whereas each case has to be considered
individually. Both in nominal use is annotated as a personal pronoun.
(2.a) the pizza
(2.b) both pizzas
(2.c) that pizza, this pizza
(2.d) his pizza
(2.e) John's pizza
anaphora-type
anaphora-type
type of anaphoric relation
none (default) ¯rst-mention of a referent
anaphoric anaphoric relation
Proper names and titles
Typical instances of proper names are geographic places (Philadelphia), persons (Judge Jenkins),
companies (Morgan Stanley & Co.), newspaper titles (The New York Times), political, social
or ¯nancial institution names (Congress, European Investment Bank). Proper names can include
noun modi¯ers or be heads of a de¯nite or inde¯nite description. In this case, the whole description
has to be marked up, not just the head.
(5.a) Berthold Brecht (full name)
(5.b) Bert Brecht (reduced full name)
(5.c) Brecht (surname)
(5.d) Berthold (¯rst name)
(5.e) Bert (nickname)
(5.f) BB (abbreviated)
(5.g) the well-known Brecht (name, modi¯ed by a de¯nite description)
(5.h) Brecht, who is author of the "Dreigroschenoper" (proper name + clause)
(5.i) the well-known Brecht (name, modi¯ed by a de¯nite description)
(5.j) Brecht, author of the "Dreigroschenoper" (proper name + apposition)
Titles in conjunction with proper names (Mr./Ms./Dr./President/Chairman) or standing alone
are treated like proper names as well, e.g.
(6) SchrÄoders...Fischerf ... Die anfÄangliche Ä Uberreaktion von Kanzlers und Au¼enministerf ...
Here, Kanzler and Au¼enminister have to be annotated as primary markables, because proper
names are inherently de¯nite and thus fall under the de¯nition of primary markables.
Parts of complex proper names cannot be analyzed separately. Proper names are primitive. So,
in the following example, Petrie in [of Petrie Stores Corp.] should not be annotated!
(7) [Milton Petrie, chairman [of Petrie Stores Corp.] said...
Note that if a proper noun is not a head of an NP, the NP is annotated as de¯nite or inde¯nite
respectively.
(8) the river Yukon - [Yukon] is the head. Yukon is a proper name, so the whole phrase is
annotated as proper name.
(9) the Yukon o±ce - [o±ce] is the head, o±ce is not a proper name, so it has to be annotated
as a de¯nite NP.
The scheme distinguishes between primary and secondary markables. Primary markables are
potential anaphors that are always subject to annotation. Secondary markables are only annotated if they happen to serve
as antecedents for primary markables.
² personal pronouns and demonstrative pronouns
² de¯nite and possessive descriptions
² proper names and titles
² pronominal adverbs
Coordination of primary markables, e.g. Bush and Putin is also a primary markable! 1
(1) [[Bush] and [Putin]]
1
1
1
1
1
1
1
referring
referring a discourse entity which can be interpreted on the basis of the previous
context
grammatical-role
grammatical-role
We use surface-oriented de¯nitions of grammatical roles. So, if used in the position of predicate,
an NP is annotated as a subject.
(43) NBI also said it has hired Prudential-Bache Securities Inc. as its ¯nancial adviser.
(44) Brazil and Venezuela are the only two countries that haven't completed steel talks
with the U.S.
referentiality
Markables are textual expressions between which anaphoric relations can be established. Only
nominal expressions are considered in the core scheme.
Generally, prototypical markables are nominal descriptions, including prepositional phrases (PPs)
and noun phrases (NP).
Size of markables
The boundaries of a markable are de¯ned according to the Maximum Size Principle.
Principle 2 (Maximum size principle) Markables of maximum size are subject to annotation.
That is:
1. one markable includes all attributes of its head noun: relative clauses, appositions, modi¯ers
and complements, right and left dislocations { marked bold in the examples below,
2. coordinated NPs/PPs are double-annotated: 1) members of a coordinated NP/PP separately
and 2) the whole coordinated NP/PP as one markable.
(19.a) [The car that went through his garden wall ] ... (relative clause)
(19.b) [The pattern of industrial development in the US] ... (NP with modi¯ers)
(19.c) [Jerald Lefcourt, a criminal defence attorney] ... (apposition)
(19.d) [That picture of a frog]p1 , where is [it]p2 (left dislocation)
(19.e) I think [he]d1's getting hooked on the taste of Vaseline, [that dog]d2. (right disloca-
tion/afterthought)
(19.f) [[SchrÄoder] und [Fischer]]
If a possessive pronoun occurs in a construction like "head noun + participle/relative clause/apposition",
we establish a secondary markable containing the controller (of a relative clause, a modi¯er or a
complement) alone. So, [his] below refers not to the whole description [(Judge Jenkins), now
known in his courthouse], but only to the secondary markable (Judge Jenkins). In case of further
references to the referent (Judge Jenkins), the whole phrase (i.e. of maximum size) - [(Judge
Jenkins), now known in his courthouse as Shake'Em Down Jenkins] - has to be annotated as
antecedent.
(20) "My belief is always, if you've got a settlement, you read it into the record," says [(Judge
Jenkins)j , now known [in [[his]j courthouse]] as [Shake'Em Down Jenkins]].
Discontinuous markables
A markable does not necessarily consist of contiguous elements. As an example, consider split-NP
constructions in German.
(24)
[BÄucher]b2
books
hat
has
Anna
Anna
[drei]b2
three
"Anna has three books./As for books, Anna has three of them"
(split-NP).
(25) You'll meet [a man]m1 tomorrow [carrying a heavy parcel]m2 (Quirk et al. 2003:930)
An important example of discontinuous markables is "group-references", i.e. to several not con-
tiguous elements, see below.
Elements that are not to be annotated on functional grounds, but share grammatical characteristics of Markables.
Do NOT annotate
² expletive expressions
(10) Then, when it would have been easier to resist them, nothing was done
(expletive it).
² Es-pronouns, pronominal adverbs, which are controllers of relative clauses
(11) Dazu kommt, dass in Werder am 24. Februar ein BÄurgermeister gewÄahlt wird und es
bisher als sicher galt, dass CDU-Amtsinhaber Werner GrÄo¼e unangefochten bleibt.
Dazu...dass, es...dass should not be annotated as markables (Dazu and es are controllers of
relative clauses).
² primary markables in idioms and collocations
(12) It sent Kate into the pits when she learned from her "friend" Martha, who seemed
to get o® on laying bad trips on people, that Harvey was getting it on with Carol. [?,
p.265]
According to Gibbs (??), we ¯nd several idiomatic phrases in this example, some of which
contain pronouns or full NPs { potential primary markables.
However, they should not be annotated as such, e.g. into the pits meaning "to be depressed",
get it on meaning "having sexual relations", neither the pits nor it can be referred to.
Note that we consider only conventiolized idiomatic expressions as idioms in our sense, i.e.
markables within productive metaphors are annotated as usual, e.g. das schwimmende City-
Schi® - a metaphor that occurred and can only be understood with respect to a speci¯c
text.
² pronominal adverbs functioning as discourse markers
(13.a) Ich habe dich angesprochen, damit du mir zuhÄorst.
"I am talking to you to let you know that you must listen to me."
(13.b) Ich habe dir das gesagt, damit du wei¼t, dass du mir zuhÄoren sollst.
(13.c) Ich habe dir das gesagt, dass du wei¼t, dass du mir zuhÄoren sollst.
² relative pronouns
Relative pronouns are annotated together with the whole relative clause it triggers as one
single markable (cf. [The car that went through his garden wall]...). If a form cannot be
unambiguously classi¯ed as a relative pronoun, apply the following test: it is a relative
pronoun if it can be substituted by "which" respectively "welch" in German. However,
relative pronouns in possessive constructions (i.e. for which the test for relative pronouns
fails) are annotated as possessive pronouns (see possessive NPs, p. 6).
(14) Und so schielen die Israelis nach Washington, an dessen /*welchem Tropf sie wirtschaftlich
und militÄarisch hÄangen,...
cf. Und so schielen die Israelis nach Washington, das/welches sie wirtschaftlich und
militÄarisch unterstÄutzt (das is a relative pronoun).
Alternatively, the following test can be applied: substitute a pronoun in question with a
possessive construction. If it works, you have a possessive pronoun, not a relative one.
(15) die Frau und deren Kinder = die Frau und ihre Kinder
The annotation is as follows in this case:
(16) Und so schielen [die Israelis]i [(nach Washington)w, [an [dessen]w Tropf] [sie]i wirtschaftlich
und militÄarisch hÄangen]w0 ,...
² prepositional phrases with prepositions as, than, bis, als, wie (in German)
Such prases are annotated as normal NPs, i.e. bis and als are not included. 3
² nominal premodi¯ers in compound nouns
(17) peanut butter, airline analyst, the creditors commettee, investment bank
Peanut, airline, cretitors and investment are no separate markables. Note that in the credi-
tor's opinion, the creditor's is annotated as a markable, since it is a nominal in genitive and
thus not a part of a compounds (see also 1.2.2, 1.2.3).
1
1
attributes of markables, note that also attributes of the relation are annotated in MMAX as attributes of markables
Only anaphoric relations have to be annotated. The de¯nition of anaphora used here conforms to
the one used in MUC-based schemes: it is "identity-of-reference direct nominal anaphora, which
can be regarded as the class of single-document identity coreference" [?].
Test: to ¯nd out, if two nominal descriptions are coreferent, try to substitute them with each
other. Note that the anaphoric relation is inherently transitive, thus every previous coreferent
markable has to be compatible with this substitution as well.
(29) Als 1999 die im Rahmen der Dorferneuerung neu gestaltete [Radeweger]r Ablage inklusive
SeebrÄucke mit viel Pomp eingeweiht wurde... Doch mit der NachrÄustung tut sich [Radewege]r0
schwer... Zu teuer, zu hÄasslich sei die Anlage, sagen die MeinungsfÄuhrer [im Gemeinderat]g
In this example, [Gemeinderat] could be coreferent with [Radewege]r0 . Although both are ex-
changeable by means of metonymy, substitution test fails for [Radewege]r, since neu gestaltete
Gemeinderatsablage is not appropriate in that context.
If a relation other than identity holds between a primary markable and its antecedent, annotate
it as follows:
² the value of referentiality feature is referring;
² BUT: the value of relation feature is non (rather than anaphoric)!
Cases of metonymy in text should also be annotated as anaphoric relations if the identity relation
between the referents in a text retains (metonymy is substituting a word for another word closely
associated with it): the State Department said... - the Stated Department o±cials claimed....
2.1.1 Choice of the antecedent
For annotation of antecedents, the following principle is to follow:
Principle 4 (Chain Principle) Any anaphoric markable has no more than one antecedent.
Mark the most recent (i.e. right-most) previous referent's mention as antecedent; all mentions
of the same referent make up an ordered chain.
preference
right+previous > left+previous
(30.a) [Die einstige Fu¼ball-Weltmacht]d zittert vor einem Winzling.
(30.b) Mit seinem Tor zum 1 : 0 fÄur die Ukraine stÄurzte der 1, 62 Meter gro¼e Gennadi Subow
[die deutsche Nationalelf]d¡elf vorÄubergehend in ein Trauma.
(30.c) Je kleiner die Kicker? daherkommen, desto grÄo¼er wird der Gegner geredet...
In this example, an anaphoric relation between die Kicker and Fu¼ball-Weltmacht is possible,
established through the relation of metonymy. But, according to the chain principle, the choice of
the right-most (i.e. most previous) antecedent is enforced. So, the antecedent of die Kicker is die
deutsche Nationalelf (ambig-ante).
Possessive descriptions
(3.a) his house
(3.b) the old man's house
Note, that the possessor must be a primary markable, too: [[his] house], [[the old man's] house].
Descriptions with a genitive attribution are regarded as possessive i®. a de¯nite genitive attribution
replaces the determiner, except for of -constructions (cf. ex. 4b).
(4.a) the old man's house (de¯nite possessive description, cf. his house and *the the old
man's house)
(4.b) the house of the old man (de¯nite non-possessive description: determiner in the house
relates to the house itself and not to its possessor)
(4.c) an old's man house(this is not a primary markable - the possessor is inde¯nite)
Syntactic cataphora
(32) Through [his] lawyers, [Mr. Antar] has denied allegations in the SEC suit ...
Syntactic cataphors are to be annotated like anaphoric links, that is, by means of a pointing
relation, but with reverse direction (from left to right). Assign the feature referring in category
referentiality (sec. 3.1).
The following examples (a nominal head followed by a restrictive modi¯er), although traditionally
classi¯ed as cataphora, should NOT be annotated as such.
(33) ... [the car that went through his garden wall ]...
(34) ... [the patterns of industrial development in the U.S]....
In case of doubt between syntactic cataphora or anaphora, decision has to be made as follows.
Principle 5 (Cataphora at the sentence level) If antecedent can be found to the right of
the anaphor, however, in the same sentence the anaphor belongs to, this antecedent should
be preferred to the right-most (i.e. the closet to an anaphor) candidate antecedent in the previous
discourse.
preference
reference to the right in same-sentence > Chain Principle (right+previous > left+previous)
(35.a) Die einstige Fu¼ball-Weltmacht zittert [vor einem Winzling]s.
(35.b) [Mit [seinem]s Tor zum 1:0 fÄur die Ukraine] stÄurzte [der 1,62 Meter gro¼e Gennadi
Subow]s [die deutsche Nationalelf] vorÄubergehend in ein Trauma.
In the example, seinem refers to Gennadi Subow who was introduced in the very ¯rst sentence as
vor einem Winzling. Following the preferences, we establish an anaphoric (cataphoric) link to the
right. Thus, the anaphoric chain looks as follows:
² seinem ! Gennadi Subow (same-sentence)
² Gennadi Subow ! vor einem Winzling (right+previous, Chain Principle)
Group-references
Reference to groups can be carried out by means of plural pronouns (e.g. they) and plural NPs,
including quanti¯ed NPs (e.g. both-NPs), which can have separate members of these groups as
antecedents. So, antecedents of plural pronouns can be non-contiguous, i.e. mentioned in di®erent
parts of a sentence or in separate clauses.
(26) [[Montedison]m]c1 now owns about 72% of [[Erbamont's]e]c2 shares outstanding.
[The companies]c said the accord was unanimously approved by a special committee of [Erbamont]e
directors una±liated with [Montedison]m.
Such groups can subsume markables with di®erent properties (e.g. grammatical form and role).
For these cases, we suggest an additional layer besides primary and secondary markables named
groups. Groups can serve as antecedents of nominal markables (but not as anaphors).
Thus, group references are annotated as follows:
1. annotate separate elements (members of the group, here Montedison and Erbamont) as
primary (or secondary) markables as usual,
2. annotate the ¯rst element (Montedison) as a group markable,
3. append the second element (Erbamont) to the ¯rst one, thus merging both elements into a
"discontinuous markable".
We suggest the primacy of referent's identity to the recency of mention. In case individual referent's
mentions have been annotated as a group somewhere, the reference is established to this group
rather than to individual mentions even right-most (i.e. the closest to an anaphor). A referent
of a group, although often being the sum of its parts, is not the same as a referent of one of its
parts. So, in the example below, which represents a sample referential chain, they refers to both,
rather than to individual mentions of Sharon and Arafat, although these individual mentions are
more recent and thus closer to the anaphor (they).
preference
identity > recency
(27) Sharons... Arafata... boths+a ... Sharons ... Arafata ... theys+a...
Besides this, the principle proposed on page 15 can be extended as follows.
Principle 3 (Primacy of markables as antecedents (extended)) If there are primary or sec-
ondary markables that can be antecedents of a markable, establish a link to these, rather than
declaring group markables, even if a primary or secondary markable is located further to the left
from the anaphor compared to the possible group markable.
preference
primary markable > secondary markable > groups
Border-line case:
(28) Now only three of the 12 judges - [[Pauline Newman]n, ([Chief Judge Howard T. Markey,
68]m)two1, and ([Giles Rich, 85]r)two2 - have patent law backgrounds]. [The latter two]two
and [Judge Daniel M. Friedman, 73]f , are approaching senior status or retirement.
Although Chief Judge Howard T. Markey, 68 and Giles Rich, 85 occur within the one and same
sentence, the conjunction of them is not a separate markable, but just a part of a larger phrase.
np-form
np-form
The feature stands for the surface structure of markables
none (default)
ne proper names
def-np de¯nite NP
*indef-np inde¯nite NP
pper personal pronouns
ppos possessive pronouns,
pds demonstrative pronouns
padv pronominal adverbs
other for special purposes, we leave this option for later extensions, choosing other
enforces to add a comment describing the type of description
Note that np-form has to be annotated both for NPs and PPs!
Discourse cataphora (anaphora of anticipation)
Discourse cataphora is a label used for non-pronominal reference forward. Sometimes an author
introduces a discourse referent by means of an underspeci¯ed NP, i.e. an NP that cannot be
interpreted only on the basis of the reader's knowledge up to this point. This way the author
tries to encourage the reader to continue reading, in order to catch up the missing information. In
the example below, die einstige Fu¼ball-Weltmacht and vor einem Winzling should be annotated
as discourse cataphors, since their referents cannot be identi¯ed until introduced explicitly in the
following text (Deutschland and Ukraine correspondingly).
(31) Die einstige Fu¼ball-Weltmacht zittert vor einem Winzling (newspaper article title)
In case one goes on reading the text, it becomes clear that die einstige Fu¼ball-Weltmacht refers
to Germany, whereas ein Winzling refers either to the Ukraine or the 1.62 meter tall ukranian footballer who made the most impact in the match 4. Discourse cataphors have to be annotated
as normal anaphors, i.e. in accordance with the Chain Principle (p. 16), i.e. the most recent
referent mention to the left (if any) is considered to be an antecedent.
phrase-type
phrase-type
np (default) noun phrase (NP)
pp prepositional phrase (PP)
other phrase type which cannot be unambiguously classi¯ed
def-np
def-np de¯nite NP
np-form
DIR-OBJ direct object: is a noun or pronoun (in Accusative) that receives the action of a verb
or shows the result of the action. It answers the question "What?" or "Whom?" after an
action verb (a transitive verb).
DIR-OBJ
other for special purposes, we leave this option for later extensions, choosing other
enforces to add a comment describing the type of description
np-form
other
no
dir reference into or within direct speech
dir
INDIR-OBJ indirect object: precedes the direct object and tells to whom or for whom the action of
the verb is done and who is receiving the direct object. There must be a direct object to have
an indirect object. Indirect objects are usually found with verbs of giving or communicating
like give, bring, tell, show, take, or o®er. An indirect object is always a noun or pronoun
(Dative in German) which is not part of a prepositional phrase.
INDIR-OBJ
yes
np
np (default) noun phrase (NP)
none
none (default)
text-level (default) reference on the text level, i.e. reference NOT into or within
the quoted material
text-level
other phrase type which cannot be unambiguously classi¯ed
other
pds
pds demonstrative pronouns
indir
indir reference into or within indirect speech
np-form
ppos
ppos possessive pronouns,
discourse-cataphora refers to a new entity, introduced into a discourse by means of
an expression with underspeci¯ed denotation ("scene-preparation" e®ect, p. 9).
Syntactic cataphors are not included here, see sec. 2.2.2.
Discourse cataphora (anaphora of anticipation)
Discourse cataphora is a label used for non-pronominal reference forward. Sometimes an author
introduces a discourse referent by means of an underspeci¯ed NP, i.e. an NP that cannot be
interpreted only on the basis of the reader's knowledge up to this point. This way the author
tries to encourage the reader to continue reading, in order to catch up the missing information. In
the example below, die einstige Fu¼ball-Weltmacht and vor einem Winzling should be annotated
as discourse cataphors, since their referents cannot be identi¯ed until introduced explicitly in the
following text (Deutschland and Ukraine correspondingly).
(31) Die einstige Fu¼ball-Weltmacht zittert vor einem Winzling (newspaper article title)
In case one goes on reading the text, it becomes clear that die einstige Fu¼ball-Weltmacht refers
to Germany, whereas ein Winzling refers either to the Ukraine or the 1.62 meter tall ukranian footballer who made the most impact in the match 4. Discourse cataphors have to be annotated
as normal anaphors, i.e. in accordance with the Chain Principle (p. 16), i.e. the most recent
referent mention to the left (if any) is considered to be an antecedent.
discourse-cataphora
anaphoric anaphoric relation
anaphoric
indef-np
*indef-np inde¯nite NP
np-form
pper
pper personal pronouns
none (default) first-mention of a referent
none
J subject: the part of the sentence or clause about which something is being said. It is usually
the doer of the action. It is normally a noun or a pronoun (in Nominative).
SBJ
other PPs and embedded elements
other
not specified
not specified (default) no decision has been made in the course of annotation
padv
np-form
padv pronominal adverbs
discourse-new a discourse entity mentioned for the ¯rst time
discourse-new
other
other annotator cannot decide how to classify
Also, instances of generic and predicative descriptions or groups whose members have
di®erent referentiality features should be marked as other. The term generic denotes
a special usage of a referring expression, such that not a particular individual or object
is meant, but rather a class of entities or features of this class.
Typical instances include generic descriptions such as the lion in example 38a or der
PrÄasident in 38b.
(38.a) The lion is an African carnivore.
(38.b) Der PrÄasident wurde immer schon durch die Stimmenmehrheit bestimmt.
'The president has always been elected by the majority of votes.'
Other cases to be classi¯ed as other are predicative descriptions.
(39) Nicht, dass beide eine Mehrheit fÄur ihre Koalition suchten, war [das Ä Argerliche
in den vergangenen Tagen] ... (predicative description)
(40) Und das ist [das Dilemma der Regierenden] (predicative description, probably
generic)
If a markable refers to something in text that cannot be declared as a primary/secondary markable
(e.g. reference to a sentence), select a referring value of referentiality feature. But do NOT
assign an antecedent to it.
pp
pp prepositional phrase (PP)
Prepositional phrases
Prepositional phrases consist of a preposition and a complement, most typically in the
form of a noun phrase. The typical prepositional phrase may indeed be viewed as a
noun phrase extended by a link showing its connection to surrounding structures. [?,
p.103]
A similar linking function is ful¯lled by case marking, too, thus we consider prepositional phrases
and noun phrases to be instances of the same underlying category of markables.
We try to keep our scheme language-independent. Due to language-speci¯c variation in the use of
prepositional phrases or in°ectional morphology we decided to provide a de¯nition that subsumes
both extremes under one label. Thus, whenever a NP appears as the complement of a PP, the
whole PP is assigned the status of a markable, not the embedded NP itself.
Note that the following prepositions and nominal conjunctions, as mentioned in the previous
subsection, should NOT be annotated:
² bis: Bis zum nÄachsten Schachzug
² als: Als 1999 die im Rahmen der Dorferneuerung gestaltete Radweger Ablage...eingeweiht
wurde, ...
In our scheme, this principle is applied to both English and German. However, morphosyntactical
motivation is primarily from German, where compound forms (pronominal adverbs) and in°ected
de¯nite prepositions (am Fenster) exist, in which the preposition has become an integral part of
the referring expression.
(18) in the period, in Japan, at its factories, of the preferred stock (preposition plus a noun phrase)
Pronominal adverbs are included in the class of primary markables, too. A list for German
pronominal adverbs in provided in Appendix in ¯g. 3.
Postpositions are treated like prepositions, i.e. such phrases as two years ago, von Beruf wegen
are prepositional phrases.
np-form
ne proper names
ne