Proceedings of the Third ACL-SIGSEM Workshop on Prepositions, pages 29–36,
Trento, Italy, April 2006. c©2006 Association for Computational Linguistics
Semantic Interpretation of Prepositions for NLP Applications
Sven Hartrumpf Hermann Helbig Rainer Osswald
Intelligent Information and Communication Systems (IICS)
University of Hagen (FernUniversit¨at in Hagen)
58084 Hagen, Germany
Firstname.Lastname@fernuni-hagen.de
Abstract
The proper interpretation of prepositions
is an important issue for automatic natu-
ral language understanding. We present
an approach towards PP interpretation as
part of a natural language understanding
system which has been successfully em-
ployed in various NLP tasks for informa-
tion retrieval and question answering. Our
approach is based on the so-called Multi-
Net paradigm, a knowledge representation
formalism especially designed for the rep-
resentation of natural language semantics.
The paper describes how the information
about the semantic interpretation of PPs is
represented in the lexicon and in PP inter-
pretation rules and how this information
is used during semantic analysis. More-
over, we report on experiments that eval-
uate the impact of using this information
about PP interpretation on the CLEF ques-
tion answering task.
1 Introduction
Advanced NLP applications such as question an-
swering require deep semantic interpretation. In
this context, prepositions play an important role
since they encode relational information. The
proper semantic analysis of prepositional phrases
is faced with various problems: (1) There is the
well-known problem of attachment ambiguities.
(2) Prepositions are highly polysemous, i.e. their
interpretation is typically context dependent. (3)
Prepositions often occur in collocations, where
their interpretation is irregular.
Although a large amount of work in the NLP
community has focused on resolving attachment
ambiguities, there are only first steps towards
a systematic description of preposition seman-
tics which has sufficient coverage for NLP appli-
cations (Litkowski and Hargraves, 2005; Saint-
Dizier, 2005). The automatic interpretation of
prepositions in English has been tackled, for ex-
ample, by Litkowski (2002), who presents hand-
crafted disambiguation rules, and O’Hara and
Wiebe (2003), who propose a statistical approach
based on collocations. However, in order to be ap-
plicable for semantic inference, the representation
of preposition semantics should ideally be inte-
grated within a full-fledged knowledge represen-
tation formalism.
In spite of the broad linguistic investigations
on preposition semantics,1 the corresponding re-
sults have seldom found their way into real NLP
applications. Information retrieval systems, on
the other hand, which claim to use NLP tech-
niques often do not cope with the semantic con-
tent of prepositions at all (even if they bear the
term semantic in their title, as with Latent Se-
mantic Analysis (Letsche and Berry, 1997)). In
many cases prepositions are even dropped as stop
words in such systems. If one really wants to
syntactico-semantically analyze texts and derive
formal semantic representations, the interpretation
of prepositions and especially the disambiguation
of their different readings is a central problem
in Indoeuropean languages like English, French,
Russian, and German.
In this paper we describe the semantic treat-
mentofthisproblemforGerman,usingtheknowl-
edge representation formalism of Multilayered
Extended Semantic Networks (MultiNet) (Helbig,
2006). The advantage of this approach is its appli-
cability to different languages and different pro-
cesses of automatic natural language understand-
ing. Since MultiNet complies with the criteria
of universality, homogeneity, and interoperability
(Helbig, 2006, Chapter 1), it can be used to for-
malize the semantics of lexemes (Hartrumpf et al.,
1See also the overview given in (Zelinski-Wibbelt, 1993).
29
Table 1: Classes of PP interpretation
complement adjunct
regular She lives in
Berlin.
They met in Au-
gust.
irregular He believes in
destiny.
He was killed on
the run.
2003) as well as that of sentences and texts (see
(Leveling and Hartrumpf, 2005), where Multi-
Net has been employed for semantic annotation of
large corpora). It can also be used as a seman-
tic interlingua throughout all NLP modules and all
applications of an NLP system (Leveling, 2005).
Typical applications that can profit from a precise
PP interpretation component are question answer-
ing (QA) systems and natural language interfaces.
2 General Classes of PP Interpretation
The semantic interpretation of prepositions in
NLPhastodealwiththefollowingtwoorthogonal
phenomena: regular (or compositional or produc-
tive) vs. irregular (or non-compositional or col-
locative) uses of prepositions and uses in comple-
ments vs. uses in adjuncts (putting aside the well-
known issue of borderline cases).2 There are thus
the four cases indicated in Table 1. In German,
PPs can occur also as invariant syntagmas in light
verb constructions (‘Funktionsverbgef¨uge’) such
as in Beschlag nehmen (‘to occupy’), to which the
complement-adjunct distinction does not apply. In
the following, we keep aside the interpretation of
prepositions in fixed phrases, which is the case
for the light verb constructions just mentioned and
also for the irregular adjunct interpretation shown
at the lower right of Table 1. This leaves us with
three types of PP interpretation: (regular) adjunct
interpretation, regular complement interpretation,
and irregular complement interpretation.
The standard examples for regular adjunct in-
terpretation are local (or directional) and temporal
PPs, which can be attached to verbs, nouns, and
adjectives:
(1) a. He walked to the museum.
b. They met in August.
(2) a. the building on the hill
b. the debate on Wednesday
2See also (Rauh, 1993).
(3) a. das
the
in
in
Europa
Europe
beliebte
popular
Spiel
game
‘the game that is popular in Europe’
b. das
the
im
in
Winter
winter
dunkle
dark
Haus
house
‘the house that is dark in winter’
Notice that German, unlike English, allows ad-
jectives with PP adjuncts (or complements) as at-
tributes of nouns, witness (3).
It is characteristic of regular adjunct interpre-
tation that the preposition has a meaning of its
own and expresses some sort of relationship. Be-
sides local and temporal specifications, there are
of course many other relationships expressed by
regular PP adjuncts such as instrumental (4-a),
comitative (4-b), and part-whole (4-c) interpreta-
tions.3
(4) a. John cleaned the floor with his shirt.
b. Mary visited London with her sister.
c. a building with large windows
The examples in (4) furthermore illustrate the well
known fact that prepositions are highly polyse-
mous in general. Within our approach described
in Sect. 4, prepositions currently have up to six-
teen readings.4
We speak of a regular complement interpreta-
tion if the PP is subcategorized and its interpre-
tation is identical to the (correct) interpretation of
thePPwhenanalyzedasanadjunct. Theseadjunct
interpretations of PPs are defined by the set of PP
rules, which are explained in Sect. 4.2. Exam-
ples of regular complements are wohnen in/auf/...
(‘to live in/on/...’), schicken nach/in/... (‘to
send to/into/...’), mitkommen mit (‘to come along
with’), and Einstieg in (‘getting in’). Here, the
choice of the preposition in the PP complement of
the lexeme is determined by the semantic charac-
terization of the complement.
In the case of irregular complement interpreta-
tion, in contrast, the selection of the preposition
is an idiosyncratic property of the subcategorizing
lexical entry. The preposition alone can be viewed
as semantically empty; only the combination of
3In our approach, 91 different relations occur, often in
combinations; see Sect. 4.2.
4Theaveragenumberofreadingsis2.44, whichisslightly
higher than the polysemy degree 2.27 (= 847/373) reported
by Litkowski and Hargraves (2005) for English. But our lexi-
concurrentlycontainsonlyfewphrasal(orcomplex)preposi-
tions like in Anbetracht (‘in view of’), which often have only
one reading.
30
the lexeme and the preposition bears semantics.
Examples for verbs, adjectives, and nouns of this
sort are glauben an (‘to belief in’), sich verlassen
auf (‘to depend on’), gut in (‘good at’), and Wut
auf (‘anger at’).
It should be noted, however, that there is a
whole spectrum of subregular phenomena within
what we called “irregular” complement seman-
tics.5 Consider, for instance, the verbs ernennen
zu, bestimmen zu (‘appoint’, ‘designate’), k¨uren
zu (‘elect’), and weihen zu (‘ordain’). Even if the
preposition zu (‘to’) could be said to semantically
express some sort of abstract goal in these cases,
anadequateinterpretationruleassociatedwiththat
preposition would have to make reference to a
rather restricted semantic class of verbs. We re-
gard it therefore as a matter of lexical semantic
organization to capturesuch subregularities within
theinterpretationofprepositionalcomplementsby
means of appropriate semantic verb templates in
the lexicon; see (Osswald et al., 2006) for details.
In the context of the present paper, an interpreta-
tion of a prepositional complement is called irreg-
ular if the interpretation is not covered by one of
our PP interpretation rules.
Table 2 show the frequency of adjunct and com-
plement interpretations in different corpora. The
numbers in the token rows are derived from auto-
matic corpus parses (see below), so there is some
noise to be expected, but the trends should be
valid.
3 The Semantic Formalism MultiNet
MultiNet is one of the few knowledge represen-
tation paradigms which have also been used as
a semantic interlingua in real-life NLP applica-
tions (Leveling and Helbig, 2002). The Multi-
Net formalism represents meanings of natural lan-
guage expressions by means of (partial) semantic
networks. A semantic network consists of nodes
representing concepts and edges representing re-
lations between concepts. Every node is addition-
ally labeled by a sort arising from an ontologi-
cally or epistemically motivated classification of
concepts (see Appendix, Table 4). Apart from
that, every node is embedded in a system of layer
attributes and their values expressing the exten-
sion type, facticity, genericity, referential deter-
mination, quantification, and others. The rela-
5See also the discussion in (Baldwin, 2005) on preposi-
tional verbs in English.
tions connecting the concepts in a semantic net-
work have to be taken from a predefined set of
expressional means, which are systematically de-
scribedandformallycharacterized(Helbig, 2006).
A strongly abbreviated description of all MultiNet
relations used in this paper can be found in Table 5
of the Appendix.
For the semantic characterization of the selec-
tional restrictions (i.e. valencies) of lexemes, an
additional set of 16 binary semantic features (such
asanimate, human, artificial, movable, andinstitu-
tion) is provided, which can be combined with the
above-mentioned sorts to yield a rich repertoire of
semantical characterizations for the description of
the slots and fillers corresponding to the valencies.
These expressional means have been used in the
computational lexicon HaGenLex (see Sect. 4.1).
4 Resources for PP Interpretation
Threesourcesofprepositioninformationareavail-
able to the syntactico-semantic parser used in our
NLP applications:6 subcategorization information
in the lexicon, context dependent PP interpretation
rules, and an annotated PP corpus.
4.1 Selection of PPs in the Lexicon
Our parser makes use of the computational lex-
icon HaGenLex (Hagen German Lexicon, see
(Hartrumpf et al., 2003)), which is a general do-
main lexicon for German with about 25,000 en-
tries (including 136 prepositions). Each entry con-
tains detailed morpho-syntactic and semantic in-
formation. In particular, the lexicon provides va-
lency frames for nouns, verbs, and adjectives (in
the lexical feature SELECT). This includes com-
plements that are syntactically realized by a PP.
Each complement is characterized by one or more
syntactic specifications and its semantic contribu-
tion to the head word. This contribution can be a
MultiNet relation (case role) or a more complex
MultiNet expression directly or indirectly con-
necting the representation of the complement and
of the head, which typically involves other com-
plements.
In order to capture semantic constraints on pos-
sible adjuncts, the set of semantic relations com-
patible with a given lexeme is specified in the lex-
icon (under the lexical feature COMPAT-R). This
information is inherited from the semantic class
6See (Hartrumpf, 2003, Chap. 3) for a description of the
parser.
31
Table 2: Class frequencies in corpora.
corpus adjunct complement regularonesamongcom-
plements (verbs only)
German QA@CLEF documents 80.6% 19.4% 17.5%
German Wikipedia (March 2005) 80.1% 19.9% —
of the lexeme so that the set of all possible ad-
junct readings for a PP (see next section) can be
filtered. Lexical entries exemplifying both aspects
are listed in Fig. 1.
4.2 Interpretation Rules for Prepositions
The second knowledge source for PP interpreta-
tion are symbolic PP interpretation rules devel-
oped for adjunct interpretations. The premise of
such a rule encodes under which semantic and
syntactic constraints a specific preposition inter-
pretation is possible; the conclusion specifies a
semantic network representing the PP semantics.
Two simplified interpretation rules are shown in
Fig. 2; the effect of the second rule is depicted in
Fig. 3. The rules are licensing possible interpreta-
tions; if several rule premises can be unified with
a given pair of a preposition’s complement and a
candidate mother, the PP disambiguation module
retreats to a statistical back-off model to resolve
this ambiguity. Currently, we have 332 rules.
The interpretation rules can be viewed as a
declarative part of the corresponding preposition
entry in the lexicon. For maintenance reasons, the
rules are stored and manipulated separately. They
are linked to the lexicon by lexeme IDs.
TherulesshowthatPPsemanticsinvolvesmany
areas of semantics. For example, MultiNet defines
around 150 relations and 91 of them are used in
the conclusion of PP rules. The 10 most frequent
ones are: LOC, VAL, TEMP, ATTR, ELMT, ATTCH,
DIRCL, INSTR, SUBM, ORIGM (see Table 5). As
exemplified by the second rule of Fig. 2, the se-
mantic network specified in the conclusion of a
rule often consists of more than one network edge;
on average, an interpretation has 1.69 edges.
4.3 Annotated PP Corpus
A third source of preposition information is an an-
notated PP corpus and statistics derived from it.
The occurrences of six frequent prepositions in
840 PPs have been manually annotated with the
correct PP attachment and most likely PP interpre-
”prahlen.1.1” [ % ‘boast’
verb
syn v-control subjeq
semsel [
v-nonment-action
select 〈
[ agt-select
sel semsel sem entity human +]
[ mcont-select
oblig −
sel syn (mit-dat-pp-syn damit-dass-syn
damit-zu-inf-syn)] 〉
compat-r {dur fin strt}
example 〈 ”(Der Mann) (prahlt) (mit seinen
Erfolgen).” 〉]]
% ‘The man boasts his successes.’
”Gegenst¨uck.1.1” [ % ‘counterpart’
n-neut
semsel [
count-n
sem [ entity prot-theor-concept
net /(attch x1 c) (equ c x0)/]
select 〈
[ nselect
sel syn zu-dat-pp-syn] 〉
example〈”DieMutteristdasGegenst¨uckzur
Schraube.” 〉]]
% ‘The nut is the counterpart to the screw.’
”unabh¨angig.1.1” [ % ‘independent’
a-nq
semsel [
semnet/(sspen1x1)(scarn1x0)(chpscn1)/
select 〈
[ aselect
sel syn (von-dat-pp-syn davon-dass-syn
davon-wh-syn)] 〉
example 〈 ”Die Manager sind nicht unab-
h¨angig von den Arbeitern.” 〉]]
% ‘The managers are not independent from
% the workers.’
Figure 1: Simplified lexicon entries for the verb
prahlen, the noun Gegenst¨uck, and the adjective
unabh¨angig.
32
aus.origm examples: eine Platte aus Kupfer (‘a plate out of copper’), ...
sort(c1)=co ∧ sort(c2)=s → (origm c1 c2)
auf.attr language examples: ein Artikel auf Englisch (‘an article in English’), ...
((sort(c1)=o ∧ info(c1)=+) ∨ sort(c1)=ad) ∧ (sub c2 ”sprache.1.1”)
→ (attr c1 c3) ∧ (val c3 c2) ∧ (sub c3 ”sprache.1.1”)
Figure 2: Examples of PP interpretation rules; c1 refers to the PP’s mother constituent, c2 to the prepo-
sition’s complement; features are explained in Sect. 3.
c2io

FACT real
QUANT one
CARD 1
ETYPE 0


SUB
c
sd15d15
c3at

FACT real
QUANT one
CARD 1
ETYPE 0


SUBc sd47d47
VAL cs
d111d111 sprache.1.1iobracketleftbiggGENER ge
ETYPE 0
bracketrightbigg
englisch.2.1iobracketleftbigg
GENER ge
ETYPE 0
bracketrightbigg
c1d∨io



FACT real
GENER sp
QUANT one
REFER indet
CARD 1
ETYPE 0




ATTR
c
c d79d79
SUBc sd47d47 artikel.1.1d∨iobracketleftbigg
GENER ge
ETYPE 0
bracketrightbigg
Figure 3: Semantic network for the noun phrase ein Artikel auf Englisch (‘an article in English’); the
relations in bold face and c3 stem from the conclusion of the PP rule auf.attr language shown in Fig. 2.
tation.7 This knowledge acts as the training set for
a machine learning component that disambiguates
attachment and interpretation of PPs (see Sect. 5).
5 Preposition Interpretation within
Semantic Parsing
All the knowledge resources described in Sect. 4
are used by the parser to determine the correct in-
terpretation of prepositions. Furthermore, PP at-
tachment ambiguities are resolved on the basis of
possible interpretations. The complement infor-
mation (valency frames) in the lexicon licenses
possible complement interpretations, the PP inter-
pretation rules (combined with the adjunct infor-
mation in the lexicon) license possible adjunct in-
terpretations. In case of alternatives, they are dis-
ambiguated using statistics derived from the anno-
tated PP corpus and a whole range of preference
scores.8
The statistical data is represented in the form of
a multidimensional back-off model. Each alterna-
tiveisdescribedbytherulename, thesemanticsof
7In case of so-called systematic ambiguities, both attach-
ments have been classified as valid. Moreover, two readings
were considered as equally likely in some cases.
8Classical rule-based approaches often apply some sort
of decision algorithm to disambiguate such cases; see e.g.
(Hirst, 1987).
the preposition’s complement, and the semantics
of the possible syntactic head. If no exact match
is found in the disambiguation statistics the num-
ber of considered alternatives and the granularity
of the description of an alternative are reduced by
backing off in these two orthogonal dimensions;
see (Hartrumpf, 2003; Hartrumpf, 1999) for de-
tails.
6 Evaluation
6.1 Intrinsic Evaluation
Experiments showed that 24.2% of verb comple-
ment interpretations are equally well produced by
adjunct rules (see column 5 in Table 3). Never-
theless the PP interpretation disambiguation task
profits from complement information because in
many of these overlap cases more than one PP in-
terpretation was possible. Also the PP attachment
disambiguation task can benefit from the comple-
ment vs. adjunct distinction because complement-
hoodisastrongindicatorofthecorrectattachment
place. A third argument for having such comple-
ment information is that it is important to model
all roles belonging to a concept on the cognitive
level; this can be easily realized by a one-to-one
correspondence between cognitive roles and com-
plements in the lexicon. Table 3 shows the number
33
Table 3: PP complements in the lexicon. Mixed PP complement means that the role can be syntactically
realized as a PP or an NP. Reg. compl. means that the complement semantics specified in the lexicon is
equally well produced by some PP interpretation rule and therefore viewed as being regular.
cat. lexemes PP compl. lexemes with PP
compl.
reg. compl. mixed PP compl. lexemes with
mixed PP compl.
v 7006 1690 1616 24.2% 105 100
n 13111 720 684 5.6% 3750 2393
of lexemes of a given category with some PP com-
plements and the total number of PP complements
forverbsandnounsinourlexicon. Thepercentage
of regular complements (24.2%) is significantly
higher than the corresponding token value in the
QA@CLEF corpus (17.5%, see Table 2). This
indicates that regular complements often have a
more optional character than irregular comple-
ments. Also in this respect, regular complements
resemble adjuncts.
The preposition interpretation method achieves
between 84% and 89% correctness for the six
prepositions supported by the hand-tagged PP cor-
pus; for prepositions without annotated corpus
data, the performance seems to drop by around 10
to 20 percent points.
6.2 Extrinsic Evaluation
One important application using the parser and the
preposition interpretation described above is In-
Sicht, a QA system for German text collections
(Hartrumpf, 2005). To measure the impact of
a deep preposition interpretation, the QA system
was run twice: with and without the PP interpre-
tation presented above. For the latter, each inter-
pretation of a PP with preposition p, an NP c2,
and syntactic head c1 was replaced by an edge
with the unique artificial relation PP.p, e.g. the
aus-PP rule in Fig. 2 would contain the conclu-
sion (PP.AUS c1 c2). The QA system was eval-
uated on the German questions from the ques-
tion answering track at CLEF (QA@CLEF) of the
years 2004 and 2005. Surprisingly, the PP in-
terpretation with unique artificial relations caused
no significant performance drop. A closer look
at all questions from QA@CLEF 2004 involving
PPs revealed that the documents with answers al-
most always contained the same prepositions as
the corresponding questions. Therefore we tried
more difficult (and often more realistic) questions
with different prepositions (and verbs or nouns).
Only natural and (nearly) equivalent paraphrases
were allowed. For the PP questions where the
QA system delivered correct answers, 14 para-
phrases were written to test the positive impact of
transforming surface prepositions to their correct
meaning.
The evaluation of PP semantics was then per-
formed using the paraphrases instead of the origi-
nal questions. For 86% of all paraphrases, the cor-
rect answer was still found when the more distant
paraphrase was used as the question for the QA
system; with the artificial relations for PPs, only
14% of the paraphrases were answered correctly.
This indicates clearly that NLP applications like
semantic QA systems benefit from a good prepo-
sition interpretation. The paraphrases that could
not be answered by the QA system with PP in-
terpretation would need more advanced reasoning
techniques to work correctly.
SomeparaphrasesinvolvedPPadjuncts. Forex-
ample, theQA@CLEFquestionqa04 055isgiven
as example (5):
(5) In
In
welchem
which
Jahr
year
wurde
was
Nelson
Nelson
Mandela
Mandela
geboren?
born?
‘InwhichyearwasNelsonMandelaborn?’
As the documents contain the correct answer in
the form of a PP with the same preposition as in
thequestion(in),shallowapproaches(andalsoour
QAsystemwiththeartificialrelations)cananswer
this question correctly. But the paraphrase (6) re-
quires that question and documents, which differ
on the surface (in-PP vs. interrogative Wann), are
transformed to the same representation (express-
ing a temporal relation).
(6) Wann
When
wurde
was
Nelson
Nelson
Mandela
Mandela
geboren?
born?
‘When was Nelson Mandela born?’
There were also some cases illustrating the im-
34
portance of a homogenous transition between the
semantics of PP adjuncts and the semantics of PP
complements. Example (7) (qa04 027) contains
an interrogative involving f¨ur, which is specified
as a complement of the verb anklagen (‘accuse’)
in the current version of HaGenLex.
(7) Wof¨ur
What-for
wurde
was
Aldrich
Aldrich
H.
H.
Ames
Ames
angeklagt?
accused?
‘For what was Aldrich H. Ames accused?’
Butinparaphraseslike(8),thesameentityappears
as an adjunct of the verb.
(8) Weswegen
What-GEN-because-of
wurde
was
Aldrich
Aldrich
H.
H.
Ames
Ames
angeklagt?
accused?
‘Because of what was Aldrich H. Ames ac-
cused?’
Thecorrectanswerisstillfoundfortheparaphrase
because both questions and the relevant document
sentences contain the same semantic representa-
tion (here, a single relation of justification).
All these paraphrases are examples of increased
recall. But also the precision of the QA system
isimprovedbecauseprepositionsensemismatches
between question and documents can be detected.
7 Conclusion and Future Work
We have presented a unified approach to the prob-
lem of automatic preposition interpretation. The
extrinsic evaluation result in the context of a QA
system encourages us to continue work in the fol-
lowing directions. The PP interpretation rules are
to be transferred to other languages. We have al-
ready started for English. The transfer on the level
of preposition readings (or PP rules) is much eas-
ier than translating prepositions. Moreover, the
coverage and quality of all three knowledge re-
sources should be further extended. For example,
we want to analyze why certain PP interpretation
rules have rarely succeeded during corpus parsing.
Extrinsic evaluations on larger and more difficult
question sets for QA systems and evaluations in
other NLP applications might help to focus further
research.

References
Timothy Baldwin. 2005. Looking for preposi-
tional verbs in corpus data. In Proc. of the 2nd
ACL-SIGSEM Workshop on Prepositions, Colch-
ester, UK.
Sven Hartrumpf, Hermann Helbig, and Rainer Oss-
wald. 2003. The semantically based computer lex-
icon HaGenLex – Structure and technological en-
vironment. Traitement automatique des langues,
44(2):81–105.
Sven Hartrumpf. 1999. Hybrid disambiguation of
prepositional phrase attachment and interpretation.
In Proc. of EMNLP/VLC-99, pages 111–120, Col-
lege Park, Maryland.
Sven Hartrumpf. 2003. Hybrid Disambiguation in
NaturalLanguageAnalysis. DerAndereVerlag,Os-
nabr¨uck, Germany.
Sven Hartrumpf. 2005. University of Hagen at
QA@CLEF 2005: Extending knowledge and deep-
ening linguistic processing for question answering.
In Carol Peters, editor, Results of CLEF 2005. Cen-
tromedia, Wien.
Hermann Helbig. 2006. Knowledge Representation
and the Semantics of Natural Language. Springer.
Graeme Hirst. 1987. Semantic Interpretation and
the Resolution of Ambiguity. Cambridge University
Press.
Todd A. Letsche and Michael W. Berry. 1997. Large-
scale information retrieval with latent semantic in-
dexing. Information Sciences, 100(1-4):105–137.
Johannes Leveling and Sven Hartrumpf. 2005. Uni-
versity of Hagen at CLEF 2004: Indexing and trans-
lating concepts for the GIRT task. In C. Peters et al.,
editors, Multilingual Information Access for Text,
Speech and Images, LNCS 3491, pages 271–282.
Springer.
Johannes Leveling and Hermann Helbig. 2002. A ro-
bust natural language interface for access to biblio-
graphic databases. In Nagib Callaos et al., editors,
Proc. of SCI 2002, volume XI, pages 133–138, Or-
lando, Florida.
Johannes Leveling. 2005. University of Hagen at
CLEF 2005: Towards a better baseline for NLP
methodsindomain-specificinformationretrieval. In
Carol Peters, editor, Results of CLEF 2005. Centro-
media, Wien.
Kenneth C. Litkowski and Orin Hargraves. 2005.
The preposition project. In Proc. of the 2nd ACL-
SIGSEM Workshop on Prepositions, Colchester,
UK.
Kenneth C. Litkowski. 2002. Digraph analysis of dic-
tionary preposition definitions. In Proc. of the ACL
Workshop on Word Sense Disambiguation, pages 9–
16, Philadelphia, Pennsylvania.
Tom O’Hara and Janyce Wiebe. 2003. Preposition se-
mantic classification via Penn Treebank and Fram-
Net. In Proc. of CoNLL-2003. Edmonton, Canada.
Rainer Osswald, Hermann Helbig, and Sven
Hartrumpf. 2006. The representation of Ger-
man prepositional verbs in a semantically based
computer lexicon. In Proc. of LREC 2006, Genoa,
Italy. To appear.
Gisa Rauh. 1993. On the grammar of lexical and
non-lexical prepositions in English. In Cornelia
Zelinski-Wibbelt, editor, The Semantics of Preposi-
tions, pages 99–150. Mouton de Gruyter.
Patrick Saint-Dizier. 2005. PrepNet: A framework
for describing prepositions: Preliminary investiga-
tion results. In Proc. of IWCS 6, Tilburg.
Cornelia Zelinski-Wibbelt, editor. 1993. The Seman-
tics of Prepositions. Mouton de Gruyter.
