Is there a way to represent metaphors in WordNets?
Insights from the Hamburg Metaphor Database
Birte L¨onneker
Institute for Romance Languages
University of Hamburg
D-20146 Hamburg, Germany
birte.loenneker@uni-hamburg.de
Abstract
This paper addresses the question whether
metaphors can be represented in Word-
Nets. For this purpose, domain-centered
data is collected from the Hamburg
Metaphor Database, an online source cre-
ated for the study of possible metaphor
representations in WordNets. Based on
the results of the analyses of French and
German corpus data and EuroWordNet,
the implementation problem is discussed.
It can be shown that a much more com-
plete representation of synsets and rela-
tions between synsets in the source do-
main as well as a clearer indication of
the level of figurativity for individual
synsets are needed before global concep-
tual metaphors can be dealt with in Word-
Nets.
1 Introduction
Based on corpus analyses of the Italian lexemes
colpire and colpo (Alonge and Castelli, 2002a)
and arrivare and separarsi (Alonge and Castelli,
2002b), Alonge and Castelli note that some impor-
tant metaphorical senses of these lexemes are not
covered by ItalWordNet, a further development of
the Italian part of the lexical database EuroWordNet
(EWN) (Vossen, 1999). They claim that metaphori-
cal data should be added to WordNets on two differ-
ent levels, or in two different ways:
1. Missing figurative senses, which turn out from
the corpus analyses, can be added directly as
new synsets (sets of semantically equivalent
words, cf. Vossen (1999, 18)).
2. “Pre-existent knowledge which constrains our
possibility to produce and/or understand novel
metaphoric expressions” (Alonge and Castelli,
2002b, 1951) should be encoded at a higher
level, because this knowledge concerns not
only single synsets, but whole conceptual do-
mains.
Alonge and Castelli work in the frame-
work of conceptual metaphors introduced by
Lakoff and Johnson (1980). They think that the
most appropriate level for representing conceptual
knowledge which underlies systematic metaphors
is a level similar to the Inter-Lingual-Index (ILI) in
EuroWordNet (EWN). This index is an unstructured
fund of concepts providing the mapping across
languages (cf. Vossen (1999, 39)). A special case
of an ILI entry is a so-called composite ILI-record,
by which regular polysemy can be covered (cf.
Vossen (1999, 40–43)). For example, the fact that
the lexeme university can refer to a building as well
as to an organization is represented by a composite
ILI, and this polysemy extends to all other lexemes
(“literals” in EWN terminology) which are members
of the respective synsets. During the construction
of EWN, language-independent composite ILIs
were used, and overgenerated polysemic relations
had to be deleted for each individual language. In
order to reduce such overgeneration, Alonge and
Castelli propose to add composite indexes similar
to composite ILIs to each individual WordNet
instead of defining general ILIs for WordNets in all
languages, and to use this composite index also for
representing regular polysemy caused by conceptual
metaphors (cf. Alonge and Castelli (2002b, 1952)).
In this paper, we present some broader anal-
yses using the current online data of the Ham-
burg Metaphor Database
1
described in (Eilts and
L¨onneker, 2002). The Hamburg Metaphor Database
(HMD) contains corpus examples in French and
German as well as synsets from EWN to which
the metaphorically used lexemes belong, and labels
for source and target domains of the metaphors in
two different naming systems. EWN synsets are
split into two groups in HMD: Those that already
have a figurative meaning in the respective WordNet
(French or German EWN), and those that contain
the metaphorically used lexeme in its literal mean-
ing. Query parameters include the synsets and the
annotated source and target domains of the metaphor
as well as the language of the corpus example. The
aim of HMD is to study the possibility of system-
atically representing information on metaphors and
potential metaphorical mappings in WordNets (cf.
Eilts and L¨onneker (2002, 100–101)).
Taking advantage of the domain based view of
HMD data, we focus on two domains represented in
the database, POLITICS and SPORTS (Section 2).
Based on the results of the analyses, we discuss
the implementation issue in more depth (Section 3).
Section 4 is the conclusion.
2 Insights from the Hamburg Metaphor
Database
This section contains analyses of material gathered
from the Hamburg Metaphor Database. Subsec-
tion 2.1 presents some main metaphorical mapping
lines in the field of POLITICS, Subsection 2.2 anal-
yses a conceptual metaphor for the target domain
SPORTS.
2.1 Mappings in the POLITICS domain
The data for the analyses are defined by those en-
tries in the Hamburg Metaphor Database that have
as their target domain 1. POLITIK (POLITICS) and
2. PARTEI (POLITICAL PARTY). The first query
yields 25 results, the second one 26.
1
http://www.rrz.uni-hamburg.de/metaphern/index en.html
[27 April 2003].
POLITICS IS FIGHT. Having a closer look at
the source domains for the target domain POLI-
TICS, we find that the main source is the FIGHT
domain (“KAMPF”), which is encoded in 20 out of
the 25 entries.
2
In what follows, we will concentrate
on the 16 examples in French language that instanti-
ate the POLITICS IS FIGHT conceptual metaphor.
Considering the synset level, we note that some
synsets are represented by more than one of their
lexemes (“literals”) in the corpus text: These are
CUcombat:2 bagar[r]e:1 bataille:4 lutte:1CV ‘the act of
fighting; any contest or struggle’ – represented by
bataille, combat and lutte in political context –,
CUtriomphe:1 victoire:1CV ‘a successful ending of a
contest’ – represented by both triomphe and vic-
toire – and CUvainqueur:1 triomphateur:1 gagnant:1CV
‘the contestant who wins the contest’ – represented
by triomphateur and vainqueur –.
3
This fact shows
that the conceptual view, which is reflected by
the synsets insofar as every synset can be seen as
a concept, is an appropriate way of approaching
metaphorical mappings.
In contrast to the results of Alonge and Castelli,
who find a lot of missing metaphorical senses in
ItalWordNet (cf. Section 1), most of the metaphori-
cally used lexemes in POLITICS IS FIGHT as col-
lected from HMD are actually represented in synsets
which already have a metaphorical sense in EWN, or
at least have been interpreted as such by HMD en-
coders. Only three lexemes out of 13, guerre ‘war’,
revanche ‘revenge’, and bataillon ‘battalion’, do not
correspond to figurative synsets. While guerre and
bataillon can be found in their “literal” meaning
in the French EWN (cf. synsets CUguerre:1CV ‘the
waging of an armed conflict against an enemy’ and
CUrevanche:1CV ‘the act of taking revenge’), batail-
lon, a collective participant of FIGHT, is completely
missing from the French EWN.
The conceptual metaphor POLITICS IS FIGHT
is only one of several possible metaphorical con-
2
In terms of the Berkeley metaphor list (Lakoff et al., 1991),
this mapping can be seen either as THEORETICAL DEBATE
IS COMPETITION or as COMPETITION IS FIGHT. If a
MENTAL EVENT (like THEORETICAL DEBATE) is a spe-
cialisation of an EVENT (like COMPETITION), then the first
metaphor can be interpreted as a specialisation of the second
one.
3
In this paper, each synset is followed by its English gloss
when it is mentioned for the first time.
ceptualisations of POLITICS. Examples of other
source domains encoded for this target in HMD
are SPORTS, THEATER, and STUDY. The POL-
ITICS IS FIGHT metaphor will thus highlight the
competitive aspects of politics: POLITICS in this
metaphor is narrowed down to POLITICAL COM-
PETITION. This COMPETING scene has physi-
cal aspects or at least physical “roots” and is an
example of a conventional metaphor, “a metaphor
we are hardly ever conscious of”, because it is one
of the ordinary ways of talking about politics (cf.
Lakoff and Johnson (1980, 5)). The metaphorically
used lexemes of the HMD example sentences and
their EWN synsets show that several elements of a
FIGHT scene, as well as the name of the event itself,
are present in the mapping:
AF Names for the whole event are CUcombat:2
bagar[r]e:1 bataille:4 lutte:1CV and CUguerre:1CV.
AF Names for events or scenes seen from
the perspective of one of the participants
are CUd´efendre:3 lutter pour:2CV ‘[argue in
defense] of causes, principles, or ideas’,
CU´ecraser:3CV ‘make ineffective’, CUperte:7CV
‘the act of losing’, CUtriomphe:1 victoire:1CV
and CUrevanche:1CV. With the exception of
CUd´efendre:3 lutter pour:2CV, which can be used
to refer to the whole action, cause or purpose
of a FIGHT as well as of POLITICS, all the
events denoted by these synsets occur only at
certain moments of the FIGHT, or of a specific
political action.
AF Finally, the metaphorical synset CUvainqueur:1
triomphateur:1 gagnant:1CV is used to refer to
one of the participants in the POLITICAL
COMPETITION, with respect to its outcome.
It turns out from this analysis that meronymy
and participant (EWN: “INVOLVED”) relations
also play a role in the mapping of this conceptual
metaphor. Unfortunately, the synsets CUguerre:1CV and
CUcombat:2 bagar[r]e:1 bataille:4 lutte:1CV, which rep-
resent the concepts of the whole competing event,
and the other involved synsets, do not display many
of these relations in the French EWN. Consulting
EWN, we find only the following non-taxonomic re-
lations (i.e. neither hyperonymic nor hyponymic re-
lations) for all the synsets mentioned above as refer-
ring to mapped aspects of the event:
AF CUguerre:1CV has mero part CUbataille:2 con-
frontation:1 conflit:1 lutte:5 combat:1CV ‘an en-
gagement fought between two military forces’
AF CUcombat:2 bagar[r]e:1 bataille:4 lutte:1CV
has mero part CUcoup:3CV ‘a powerful stroke
with the fist or weapon’
AF CUvainqueur:1 triomphateur:1 gagnant:1CV
antonym CUperdant:2 non-valeur:1CV ‘a contes-
tant who loses the contest’
AF CUtriomphe:1 victoire:1CV antonym CUd´efaite:2CV
‘an unsuccessful ending’
The stated subevents in this list are not bound to
the overall event, and no participant relations are en-
coded for any event or subevent in EWN. This lack
in relations means that the structure of the source
domain is not visible from EuroWordNet.
Figure 1: Literal and possibly figurative synsets.
action:4 acte:6
action de groupe:1
collision:2 bataille:1 lutte:2
’an open clash between two
opposing groups’
combat:2 bagar[r]e:1
bataille:4 lutte:1 [F?]
’the act of fighting; 
any contest or struggle’
coup:3 [L]
’ a powerful stroke with 
the fist or weapon’
has_mero_part
guerre:1 [L]
’the waging of
an armed conflict ...’
has_holo_part
has_holo_part
action militaire:1 action:8 [L]
’a military engagement’
lutte:3 bataille:3 [L]
’a hostile meeting
of opposing 
military forces...’
combat:1 lutte:5 
conflit:1confrontation:1 
bataille:2 [L]
’an engagement fought 
between military forces’
[L] literal
[F] figurative
has_hyponym
Furthermore, it is sometimes difficult to tell
whether a synset has an intended figurative mean-
ing or not, let alone which synsets are re-
lated by a literal-figurative relation. Consider
the conceptual-semantic relations of the synsets
CUbataille:2 confrontation:1 conflit:1 lutte:5 com-
bat:1CV and CUcombat:2 bagar[r]e:1 bataille:4 lutte:1CV
as represented in EWN, displayed in Figure 1. For
two reasons, it is sensible to believe that the second
one has a figurative meaning, as opposed to the first
one:
1. CUcombat:2 bagar[r]e:1 bataille:4 lutte:1CV is de-
fined as ‘[...];anycontest or strugggle’, which
might also refer to non-military and possibly
non-physical violence.
2. There are no other synsets containing the lex-
emes bataille and lutte for which a figurative
reading in the ‘conflict’ sense would be possi-
ble.
On the other hand, the gloss of the synset
CUcoup:3CV ‘a powerful stroke with the fist or weapon’,
which is encoded to be a part of CUcombat:2
bagar[r]e:1 bataille:4 lutte:1CV, gives a clearly literal
meaning to CUcoup:3CV, which results in an incon-
sistency between the possibly metaphorical sense
of the holonym and the strictly literal sense of the
meronym (part).
A POLITICAL PARTY IS A FAMILY. Another
conceptual metaphor that turns out to be frequent in
the politics domain, at least in German texts, is the
one in which a POLITICAL PARTY (“PARTEI”) is
viewed in terms of a FAMILY (“FAMILIE”).
4
This
conceptual metaphor is mainly a mapping of one so-
cial group to another (cf. Eilts and L¨onneker (2002,
107)). A query for these source and target domains
yields 20 results for German, which will be the sub-
ject of the following analysis.
A closer look at the synsets shows a slightly dif-
ferent picture from what we obtained in the last anal-
ysis. 19 distinct lexemes are used, but only five of
them are present in German EWN synsets. These
refer mainly to individual family members: Vater
‘father’, Bruder ‘brother’, Schwester ‘sister’, Enkel
‘grandchild’, out of which only Vater can be found
in a figurative synset (CUVater:2CV ‘a person who holds
an important or distinguished position in some orga-
nization [. . . ]’).
The low coverage of the lexemes is also due
to the current practice in HMD not to encode
parts of compounds as individual lexemes (cf.
Eilts and L¨onneker (2002, 107)). In doing so, the
following additional family member lexemes cov-
ered by the German EWN would appear: Mut-
ter ‘mother’ (cf. Mutterpartei), Vetter ‘cousin’ (cf.
Vetternwirtschaft), and three further occurrences of
4
A different conceptual source domain for POLITICAL
PARTY attested in HMD is BUILDING.
Vater (cf. Vaterfigur, Vatermord,
¨
Ubervater). The
most direct mapping of the metaphor is indeed rep-
resented in the lexeme Parteifamilie ‘family of a
political party’, which summarises the whole con-
ceptual metaphor in one word. While it would be
interesting to further discuss the topic of German
compounds, for the sake of brevity we will go on
to other aspects of the conceptual metaphor that are
concerned by the mapping.
In addition to a neutral reference to individual
family members, two other referring methods can be
detected from the metaphorically used lexemes not
encoded in the German EWN:
1. collective nouns for younger members of the
family: Nachwuchs ‘offspring’ (in the com-
pounds Nachwuchskraft, Nachwuchspolitiker
and Parteinachwuchs);
2. subconcepts or referring expressions encom-
passing specific role/character attributions:
Ziehkind ‘foster child’ and Patriarch ‘patri-
arch’.
In the German EWN, family member synsets are
not linked to the family synset (CUFamilie:1CV) itself.
It seems that with a few synset additions to the
German EWN, compound splitting, and the con-
nection of the family members to the family synset
using the has holo member relation, the mapping
of this conceptual metaphor could be implemented
more easily than the one previously discussed. How-
ever, as further elements in the mapping show, also
this social group metaphor can be extended to an
event metaphor. The lexemes Erbe (covered by the
German EWN as CUErbe:1CV ‘any acquisition from
past generations’, represented also in the compound
Erblast) and Hinterlassenschaft, which could be
added to the CUErbe:1CV synset
5
but has also other
senses, actually denote “participants” in the seman-
tic patient role of an event. The “inheritage” (in form
of qualities, power and achievements, but also neg-
ative characteristics and problems) is handed over
from one member of the political party, usually
mapped to the father of the family, to other mem-
bers, preferably younger or less experienced ones.
5
Cf. the corresponding verb synset in the German EWN:
CUvererben:1 hinterlassen:1CV.
Only this inheriting event conveys also meaning to
the position of these members in the political party.
2.2 Mappings in the SPORTS domain
SPORTS IS FIGHT. Turning to the field of
sports, we query the Hamburg Metaphor Database
for all French examples of the target domain
SPORTS (“SPORT”). In the 34 results, the list of
source domains shows that FIGHTING (“KAMPF”,
16 times), also in its special form of WAR
(“KRIEG”, 8 times), is the predominant source
metaphorically used to view this target domain.
6
The following analysis will be based on the
24 French examples instantiating either the con-
ceptual metaphor SPORTS IS FIGHT (“SPORT
IST KAMPF”) or SPORT IS WAR (“SPORT IST
KRIEG”). 21 distinct lexemes occur in the exam-
ples; only two of them are not represented in any
synset in the French EWN, the others instantiate 6
distinct synsets with literal meaning and 13 distinct
synsets with figurative meaning, following the inter-
pretation of HMD.
Even more clearly than in the POLITICS IS
FIGHT conceptual metaphor, the resulting synsets
show that not only the event itself, but several
aspects of the FIGHT event are mapped across
domains in the SPORTS IS FIGHT (containing
SPORTS IS WAR) metaphor. The examples anal-
ysed are from a football corpus and can be narrowed
down to A MATCH (as a sports event) IS A FIGHT.
The mapped aspects of the event are enumerated in
the list below. If a synset is contained in the French
EWN only in a literal meaning, we indicate this by
the addition of [L]. Note that events and subevents
can equally be expressed by verb or noun synsets.
1. There are synsets referring to the entire
FIGHT event matched to the SPORTS event:
CUcombat:2 bagar[r]e:1 bataille:4 lutte:1CV,
CUlutter:4 se battre:4CV ‘be engaged in a fight;
carry on a fight’ [L], CUguerre:1CV [L].
2. Actions of a participant or several participants
collectively are represented by CU´ecraser:3CV
6
The main mapping lines in Berkeley terms are again COM-
PETITION IS FIGHT (17 times; also as COMPETITION IS
PHYSICAL AGGRESSION (2 times), cf. “COMPETITION IS
1 ON 1 PHYSICAL AGGRESSION, Source domains: 1-on-1
physical aggression, fight. Target domain: competition”(Lakoff
et al., 1991, 66)) and COMPETITION IS WAR (8 times).
‘make ineffective’, CUd´efier:1 provoquer:1CV
“Fischer challenged Spassky to a match”,
CUmanoeuvre:1CV ‘a move made to gain a tactical
end’, CUarmer:2 consolider:3 renforcer:1CV ‘sup-
port or hold steady, as if with a brace’.
3. Singular actions occurring during the event,
in which a “hostile action” is performed not
against a person, but against an object (here:
the ball), are represented in CUcoup:5CV ‘the act of
swinging or striking at a ball [...]’,CUfrapper:7CV
‘hit a ball’.
4. The result of the event or of a subevent seen
from a participant-dependent view is present in
CUtriomphe:1 victoire:1CV and CUconquˆete:2CV ‘the
act of conquering’ [L].
5. The participants of the event can be identi-
fied by the synsets CUadversaire:2 opposant:1
partie adverse:1CV ‘a hostile person who tries
to do damage to you’, CUopposant:2 adver-
saire:1 ennemi:3CV ‘a contestant that you are
matched against’ and CUagresseur:1 attaquant:1CV
‘person who attacks someone’ [L]. For some
of the participants, collective nouns are used:
CUd´efense:10CV ‘the defensive football play-
ers[...]’
6. Finally, the instrument can be mapped:
CUarme:1CV ‘weaponry used in fighting or hunt-
ing’ [L] (here used to refer to one of the football
players and his special qualities).
In some examples, a quite radical type switching
occurs. Actions performed by at least one person,
as subevents of the sports event (match), are rep-
resented in EWN only in synsets denoting the par-
ticipants performing these actions: CUattaque:6CV ‘the
team that has the ball (or puck) and is trying to score’
(here in ˆetre `a l’attaque), CUd´efense:3CV ‘the team that
is trying to prevent the other team from scoring’
(here in faire une grosse d´efense). A place involved
in the source domain, CUcamp:4 bivouac:1CV ‘tempo-
rary living quarters, specially built by the army for
soldiers’, is used to refer to the players of a team and
only attested by camp in the corpus.
The uncovered lexemes instantiate still other as-
pects of the event. Duel ‘duel’ refers to a part of
the event from a participant-independent view, while
the remaining lexeme, offensif ‘aggressive’, charac-
terises one of the participants or his actions. Again,
with the exception of some antonymy encodings,
there are no relations in EWN binding any of these
elements together.
3 Implementation issues
Based on the results of the analyses in Sec-
tion 2, the idea of a composite index for regular
metaphorical mappings which was brought
forward by Alonge and Castelli (2002a) and
Alonge and Castelli (2002b) will be discussed. A
necessary continuation of the work by Alonge and
Castelli, who do not deal with implementation
details, is to discuss this issue in more depth.
Alonge and Castelli propose to deal with
metaphorical mappings using a kind of composite
index. This sort of index has already been in use
in form of a composite Inter-Lingual-Index during
the construction of EWN and accounted for regu-
lar metonymic polysemy (cf. Section 1). Figure 2
shows an example of an existing composite ILI from
EWN. It indicates that there is a metonymic poly-
semy of the lexeme country, which has three senses
(cf. “gloss”):
1. territory which has been singled out for some
purpose;
2. the people of a nation or country or a commu-
nity of persons bound by a common heritage;
3. the people, government and territory of a state.
The effect of the composite ILI is that each synset
identified by one of the three target ILIs contains an
eq metonym relation to the composite ILI, and this
in all languages involved in EWN. By accessing this
relation from one of the individual synsets, the user
can thus see which synsets are bound together by the
metonymic relation. However, considering figura-
tivity aspects, there is no rule defining which synset
should be the literal one, and this information cannot
be obtained accessing the synsets.
Let us consider an encoding of the conceptual
metaphor POLITICS IS FIGHT in such a compos-
ite index. As we have seen, all encoded aspects of
this metaphor are in some way related to an over-
all event. One of the most general source events
Figure 2: A composite EuroWordNet ILI.
0 ILI_RECORD
1 PART_OF_SPEECH "n"
1 ADD_ON_ID 30
1 GLOSS "territory which has been singled out for some purpose"
"the people of a nation or country or a community of persons
bound by a common heritage"
"the people, government and territory of a state"
1 VARIANTS
2 LITERAL "country"
3 SENSE 1
2 LITERAL "country"
3 SENSE 2
2 LITERAL "country"
3 SENSE 3
1 EQ_RELATION "eq_metonym"
2 TARGET_ILI
3 WORDNET_OFFSET 5400698
[={country:1 state:1 land:3 nation:1}]
3 WORDNET_OFFSET 5208026
[={nation:2 nationality:1 land:2 country:2 a people:1}]
3 WORDNET_OFFSET 5209013
[={state:1 nation:1 country:1 commonwealth:1 res publica:1
body politic:1}]
Figure 3: Metaphorical composite index entries.
[...]
1 GLOSS "an armed conflict against an enemy"
a political conflict"
1 VARIANTS
2 LITERAL "guerre"
3 SENSE 1
2 LITERAL "guerre"
3 SENSE 2
1 EQ_RELATION "eq_metaphor"
2 TARGET_ILI
3 WORDNET_OFFSET 540597
[={guerre:1}]
3 WORDNET_OFFSET ToBeCreated
[...]
1 GLOSS "an engagement fought between two military forces"
"the act of fighting; any contest or struggle"
1 VARIANTS
2 LITERAL "lutte"
3 SENSE 5
2 LITERAL "lutte"
3 SENSE 1
1 EQ_RELATION "eq_metaphor"
2 TARGET_ILI
3 WORDNET_OFFSET 535646
[={bataille:2 confrontation:1 conflit:1 lutte:5 combat:1}]
3 WORDNET_OFFSET 645833
[={lutte:1 bataille:4 bagar[r]e:1 combat:2}]
can be referred to by the French literal guerre ‘war’.
In a composite index for the conceptual metaphor
A POLITICAL CONFLICT IS A WAR it could be
stated that guerre means ‘an armed conflict against
an enemy’, but also (figuratively) ‘a political con-
flict’. The figurative synset does not yet exist in
EWN, so it should be created. A similar composite
index could be used to bind together already existing
different senses of lutte ‘fight’. The respective parts
of the composite index entries are shown in Figure 3.
This representation would ensure eq metaphor
links from all included synsets to the composite
index entry. Though, it would still not be clear
which synset bears the literal meaning. As there
could be several figurative meanings derived from
the same literal synset (for example, a figurative
meaning of guerre ‘a sports event’ could be added),
the index would not solve the problem of telling
whether a literal-figurative relation exists between
given synsets, and even whether a synset has a fig-
urative meaning or not. A convention could be es-
tablished that the first synset mentioned in the com-
posite index entry has to be the literal one. How-
ever, if the composite index exists for each language
separately, as it might be requested because of dif-
ferent use of conceptual metaphors in different lan-
guages, and if it covers only the existing literal and
metaphorical synsets, there seems to be no need to
encode the literal-figurative relation at a higher level
than the synset itself. A new conceptual-semantic
relation like derived from literal could be used as
well, and would keep track of the direction of the
mapping.
Alonge and Castelli (2002b) note that a compos-
ite index for metaphors could account for more than
one mapping and mention as an example the reg-
ular polysemy of motion verbs which can be used
to metaphorically refer to stages in a love relation-
ship. As a result of the analyses performed on data
from the Hamburg Metaphor Database and the im-
plementation possibilities presented above, it turns
out that indeed a composite index can only be useful
if it accounts for several, and ideally for all map-
pings which occur within one encompassing con-
ceptual metaphor. Only a generalisation from one
index entry to all synsets related to the source synset
will be able to cover also novel metaphorical uses of
words within existing conceptual metaphors.
As we have seen from the analyses, in most
cases the most general mapping of the conceptual
metaphor links a source event to a target event. This
mapping is the one that will have to be manually en-
coded after analyses of texts dealing with different
topics. However, for the mapping of all other as-
pects, metaphorical relations could be supposed and
– in an automated or assisted way – added to the
index. The fact of linking an entity to another by a
metaphorical index should then result in the creation
of corresponding synsets in the target domain for all
entities (hyponyms, parts, participants, etc.) that are
related to the source entity by sense relations. The
easiest way to create such a (potential) metaphorical
synset would be to copy the lexemes of the literal
synset and to add a predefined gloss.
For example, if a metaphorical relation is in-
stalled between CUguerre:1CV and the newly created
CUguerre:2CV, metaphorical synsets and relations for
all its hyponyms, parts, subevents, participants etc.,
as mentioned in the analyses, should also be cre-
ated. Standard glosses with variables for the dif-
ferent mapped events and elements would adapt to
these new synsets. For example, in the gloss
a[CUˆetre:1 mortel:1 mortel:1 homme:7CV] that par-
ticipates in a [POLITICAL COMPETITION] and
can be compared to a [CUvainqueur:1 triomphateur:1
gagnant:1CV] that participates in a[n] [ARMED
CONFLICT]
the event names are taken from the glosses of tar-
get and source event respectively, and the participant
description (its topmost hyperonym as well as its
source name) from the source synset. For more de-
tails on WordNet glosses, cf. Pala and Smrz (2002).
This method of adding a composite index for
global conceptual metaphors has its limits for dif-
ferent reasons:
AF First of all, the status of the French and German
EWN do not allow such a mapping, because
they encode only very few non-taxonomic re-
lations. Some examples from the analyses even
show that the set of encodable relations from
EWN would not be sufficient to cover the struc-
ture of all domains. Analyses of source do-
mains in the framework of semantic frames
(Fillmore and Atkins, 1994)
7
or concept frames
(L¨onneker, 2002) can help identifying needed
conceptual-semantic relationships and instanti-
ations of them.
AF A lot of figurative synsets are already present
in EWN and often differ slightly from the lit-
eral synsets in their lexemes, so that many du-
plicates would be created.
AF Automatically created index entries would con-
tain no information about whether an individual
mapping is actually attested or not.
7
Cf. also the FrameNet Database:
http://framenet.icsi.berkeley.edu [27 April 2003].
These considerations show that there is still a lot
of work to do before a representation of whole struc-
tures of conceptual metaphors can be envisaged.
Based on detailed analyses of corpora from differ-
ent domains, the coverage of source domains both
in synsets and in relations has to be improved, and
the status of individual synsets regarding their level
of figurativity has to be clarified. It can be hoped
that new tools like VisDic (Pavelek and Pala, 2002;
Horak et al., 2003) and DEB (Smrˇz and Povoln´y,
2003) for editing lexical databases will enable the
integration of independent further developments of
EWN (for example GermaNet) with EWN data and
the creation of new relationships.
4 Conclusion
The answer to the question whether metaphors can
be represented in WordNets depends on the interpre-
tation of metaphor.
With respect to EuroWordNet, if individual lex-
emes with metaphorical meaning are considered, it
can be stated that a lot of them are already repre-
sented by synsets that can or must be used figura-
tively. What is missing are relations between literal
and figurative individual synsets.
If a metaphor is considered as a structured map-
ping of one conceptual domain to another, the pro-
ductivity of the envisaged index, as well as its poten-
tial to account for new metaphorical word senses,
increases drastically. Unfortunately, so does also
the complexity of the implementation task, for ex-
ample by means of a composite metaphorical in-
dex. The analyses of two different target domains,
POLITICS and SPORTS, show that at first a much
more encompassing representation of the source do-
mains, mainly events and their characteristics like
parts, participants, actions and characteristics of the
participants, is necessary in order to represent the
domain structure. This can be achieved by encoding
more instances of EWN relations like Meronymy
and ROLE/INVOLVED, as well as possibly addi-
tional relations.
For the time being, continuing the domain cen-
tered perspective adopted by the Hamburg Metaphor
Database, the following research lines seem the most
promising:
1. More clearly identify literal and figurative
synsets in the mapped domains, and their link-
ing relation.
2. Individuate elements in source domains and the
conceptual-semantic relations between them.
Acknowledgements
I am grateful to Carina Eilts from the Hamburg
Metaphor Database for her great help. My thanks
also go to Tom´aˇs Pavelek, Pavel Smrˇz and the
anonymous reviewers for their comments.

References
Antonietta Alonge and Margherita Castelli. 2002a.
Metaphoric expressions: an analysis of data from a
corpus and the ItalWordNet database. In Proceedings
of the First Global WordNet Conference, pages 342–
350, Mysore, India.
Antonietta Alonge and Margherita Castelli. 2002b.
Which way should we go? Metaphoric expressions
in lexical resources. In Proceedings of the third Lan-
guage Resources and Evaluation Conference (LREC-
02), volume VI, pages 1948–1952, Las Palmas, Gran
Canaria.
Carina Eilts and Birte L¨onneker. 2002. The Hamburg
Metaphor Database. metaphorik.de, 03:100–110.
Charles J. Fillmore and B. T. Sue Atkins. 1994. Starting
where the dictionaries stop: The challenge of corpus
lexicography. In B. T. Sue Atkins and Antonio Zam-
polli, editors, Computational Approaches to the Lexi-
con, pages 349–393. Oxford University Press, Oxford.
Ales Horak, Karel Pala, and Pavel Smrz. 2003. Lexical
semantic networks and ontologies in XML, their view-
ing and authoring. To appear in Proceedings of the
Workshop on WWW Based Communities For Knowl-
edge Presentation, Sharing, Mining and Protection -
PSMP 2003, within CIC 2003 (International Confer-
ence on Communications in Computing), Las Vegas,
Nevada, USA.
George Lakoff and Mark Johnson. 1980. Metaphors we
live by. University of Chicago Press, Chicago/London.
George Lakoff, Jane Espenson, and Alan Schwartz.
1991. Master metaphor list, Second draft copy.
http://cogsci.berkeley.edu [27 April 2003].
Birte L¨onneker. 2002. Building concept frames based
on text corpora. In Proceedings of the 3rd Interna-
tional Conference on Language Resources and Evalu-
ation (LREC-02), volume I, pages 216–223, Las Pal-
mas, Gran Canaria, Spain.
Karel Pala and Pavel Smrz. 2002. Glosses in WordNet
1.5 and their standardization/consistency. In Proceed-
ings of the LREC Workshop on WordNet Structures and
Standardisation, and How These Affect WordNet Ap-
plications and Evaluation, pages 20–23, Las Palmas,
Gran Canaria.
Tomas Pavelek and Karel Pala. 2002. WordNet stan-
dardization from a practical point of view. In Proceed-
ings of the LREC Workshop on WordNet Structures and
Standardisation, and How These Affect WordNet Ap-
plications and Evaluation, pages 30–34, Las Palmas,
Gran Canaria.
Pavel Smrˇz and Martin Povoln´y. 2003. DEB — Dic-
tionary Editing and Browsing. In Proceedings of the
EACL-03 Workshop on Language Technology and the
Semantic Web: The 3rd Workshop on NLP and XML
(NLPXML-2003), pages 49–55, Budapest, Hungary.
Piek Vossen. 1999. EuroWordNet general doc-
ument Version 3, Final. http://www.illc.uva.nl/
EuroWordNet/docs.html [27 April 2003].
