Towards a Unified Multilingual Ontology for Rhetorical Figures
Yetian Wang
1 a
, Ramona K
¨
uhn
3 b
, Randy Allen Harris
2 c
, Jelena Mitrovi
´
c
3,4 d
and Michael Granitzer
3 e
1
David R. Cheriton School of Computer Science, University of Waterloo, 200 University Avenue West, Canada
2
Department of English Language and Literature, University of Waterloo, 200 University Avenue West, Canada
3
Faculty of Computer Science and Mathematics, University of Passau, Innstraße 43, Passau, Germany
4
Institute for Artificial Intelligence Research and Development of Serbia, Fru
ˇ
skogorska 1 21000 Novi Sad, Serbia
Keywords:
Ontology, Rhetorical Figures, Knowledge Representation, Computational Rhetoric, Language Modelling.
Abstract:
Formal ontologies for rhetorical figures have been developed to improve the computational detection for dif-
ferent applications in the area of Natural Language Processing, such as hate speech and fake news detection,
argumentation mining, and sentiment analysis. The existing ontologies all model different aspects of rhetorical
figures, thus creating a variety of formalisms and in the worst case, creating incompatibilities and contradictory
representations. In this paper, we focus on figures of perfect lexical repetition and their representation in three
ontologies in three different languages: The Ploke ontology, the Serbian RetFig, and the German GRhOOT
ontology. We combine those ontologies to benefit from synergy effects and create a multilingual, coherent,
robust, and modular ontology for rhetorical figures of perfect lexical repetition.
1 INTRODUCTION
A rhetorical figure is an extra-grammatical linguis-
tic device which generates attentional effects such as
salience, aesthetic pleasure, and a mnemonic effect at
the receiver side. Rhetorical figures are widespread
in all registers, genres, and dialects of all languages.
There simply does not exist a pure literal language
(Harris et al., 2017). Common rhetorical figures,
such as rhyme and metaphor are encountered in ev-
eryday conversations. One class of rhetorical fig-
ures is called trope, which is characterized by se-
mantics, e.g., a metaphor is the mapping of simili-
tude between semantic domains conveyed. In con-
trast, the class scheme is characterized by the form,
e.g., a rhyme is the repetition of final syllables of
words in a passage. Rhetorical figures have received
more and more attention in the field of Natural Lan-
guage Processing (NLP) in recent years, as their sig-
nificance has become clear for tasks such as argu-
mentation mining (Mitrovi
´
c et al., 2017; Lawrence
a
https://orcid.org/0000-0002-6984-7256
b
https://orcid.org/0000-0002-9750-0305
c
https://orcid.org/0000-0002-9324-1879
d
https://orcid.org/0000-0003-3220-8749
e
https://orcid.org/0000-0003-3566-5507
et al., 2017; Green and Crotts, 2020; Green, 2020),
sentiment analysis (Karp et al., 2021), fake news/hate
speech detection (Musolff, 2015; Caselli et al., 2020;
Lemmens et al., 2021), text summarization (Alli-
heedi and Di Marco, 2014), text improvement (Harris
and DiMarco, 2009), authorship attribution (Strom-
mer, 2011; Java, 2015), machine translation (Clarke,
2019), and translation with focus on maintaining the
rhythm of text (Lagutina et al., 2020). To use rhetor-
ical figures in a computational context, formal mod-
els like ontologies were developed. The goal of the
project RhetFig (Harris et al., 2017; Harris and Di-
Marco, 2009; Kelly et al., 2010) is to build a neuro-
cognitive ontology for rhetorical figures. The ontol-
ogy is ‘neuro-cognitive’ in the sense that it not only
captures ‘isA’ and ‘partOf’ relations among rhetori-
cal figures, it can also infer how a rhetorical figure
generates potential attentional and mnemonic effects,
e.g., whether a word or word group is more salient,
in order to understand rhetorical strategies embed-
ded in an utterance (Harris and DiMarco, 2009; Har-
ris et al., 2017). Rhetorical figure ontologies are re-
quired for automatic figure recognition, annotation,
and generation, which are critical for the aforemen-
tioned tasks. From RhetFig, the RetFig ontology
was developed (note: without the letter “h”) (Mlade-
novi
´
c and Mitrovi
´
c, 2013), an ontology that describes
Wang, Y., Kühn, R., Harris, R., Mitrovi
´
c, J. and Granitzer, M.
Towards a Unified Multilingual Ontology for Rhetorical Figures.
DOI: 10.5220/0011524400003335
In Proceedings of the 14th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2022) - Volume 2: KEOD, pages 117-127
ISBN: 978-989-758-614-9; ISSN: 2184-3228
Copyright
c
 2022 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
117

most of the rhetorical figures in the Serbian language.
This ontology was recently translated and adapted to
the German language, modelling 110 German figures,
and named GRhOOT (K
¨
uhn et al., 2022).
The Ploke ontology (Wang et al., 2021) focuses
on modelling the class of rhetorical figures of perfect
lexical repetition, i.e., ploke, from a neuro-cognitive
perspective. It also models how attentional effects are
generated by such rhetorical figures. For example, the
phrase “long long ago”
1
is a figure of perfect lexical
repetition with the word “long” repeated. As stated
by Harris, “[m]ore occurrences of a word increase the
scale of an evoked concept in a constrained range of
ways” (Harris, 2020). Receivers perceive this phrase
with a sense of time that is further away in history
compared to another phrase that simply states “long
ago”. The definition of ‘perfect lexical repetition’ was
adopted from Fahnestock and Harris: “the repetition
of a word or word group with no variation of signans
or signatum” (Fahnestock, 2002; Harris, 2020). In
other words, both the form and meaning of the re-
peated words or word groups stay the same. Thus,
ploke is distinguished from figures of ‘partial lexical
repetition’ such as synonymia, in which synonyms are
used to convey the same meaning, e.g., “How weary,
stale, flat, and unprofitable . . . ” (Shakespeare, Ham-
let, 1.2) (Shakespeare, 2014); polyptoton, in which
words and word groups are repeated with different
morphology, e.g., “ . . . with the remover to remove”
(Shakespeare, Sonnet 116) (Shakespeare, 2014); and
antanaclasis, in which words and word groups are re-
peated with different meanings, e.g., “we must . . . all
hang together, or . . . we shall all hang separately”
(Benjamin Franklin). In this paper, we combine the
figures of perfect lexical repetition of three ontolo-
gies (Ploke, RetFig, GRhOOT) into one multilingual
ontology.
We chose those three ontologies as we want to
focus on multilingual aspects, formal categorization,
and neuro-cognitive affinities. To the best of our
knowledge, the RetFig and GRhOOT ontologies are
the only known non-English ontologies that formally
model rhetorical figures, but they do not include the
effects or purposes of rhetorical figures. However,
GRhOOT already includes multilingualism as it is a
translation and adaption of the Serbian RetFig. We
focus on figures of perfect lexical repetition as they
are extra-grammatical and neuro-cognitively moti-
vated and are therefore language independent. The
Ploke ontology is a perfect choice, as it models fig-
ures of perfect lexical repetition in English and in-
cludes neuro-cognitive aspects. Therefore, the Ploke
1
This is an instance of a figure called epizeuxis, which
we will discuss in Section 3.2.
ontology should be able to incorporate language fea-
tures specific to the Serbian and German ontologies.
Furthermore, all three ontologies aim to be combined
with other ontologies as future work: Therefore, they
were already modelled to allow an extension or adap-
tion. We discuss how the ontologies are fit or unfit to
be connected, and the steps we take to alter any in-
compatible components in the ontologies. However,
each ontology was modelled differently and inconsis-
tencies or deviations are expected. The Ploke ontol-
ogy serves as a baseline since it models the most spe-
cific rhetorical figure concepts among the three on-
tologies.
Our contribution is that we tackle the problem that
many different ontologies for rhetorical figures are de-
veloped that are then incompatible with each other.
Merging rhetorical figure ontologies is a difficult task
due to inconsistencies not only in ontology modelling,
but also in domain-specific knowledge of rhetorical
figures. Therefore, it is critical to understand potential
inconsistencies that may occur when merging them.
By merging the ontologies, they benefit from each
other and synergy effects can be used. We take an-
other step forward to the enhancement of machine-
readable ontologies for rhetorical figures. The re-
sulting ontology is a more robust, tightly-connected
yet modular ontology (or ontological suite), which
demonstrates the re-usability of ontologies. Section
2 summarizes related work on rhetorical figure on-
tologies. Section 3 compares the ontologies, their
rhetorical figures, definitions, and structures. In Sec-
tion 4, we show the differences and inconsistencies
of those ontologies regarding terminology, modelling,
and concepts. A sketch of the combined ontology is
presented in Section 5.
2 RELATED WORK
Fahnestock provides an excellent overview of the de-
velopment of rhetorical figure classification (Fahne-
stock, 2002). As different names and definitions of
rhetorical figures exist, a classification can never be
unambiguous against the background of the inherited
terminology. With the help of ontologies, researchers
have tried to formally model rhetorical figures and
their properties. One of the first ontologies in this
direction models linguistic operations (addition, dele-
tion, etc.) and neuro-cognitive affinities (comparison,
symmetry, etc.), including a formal description of
rhetorical figures and their operations (Harris and Di-
Marco, 2009). The approach is limited to words that
do not have multiple meanings or different forms. The
RhetFig (Rhetorical Figure Ontology) Project (Harris
KEOD 2022 - 14th International Conference on Knowledge Engineering and Ontology Development
118

and DiMarco, 2009; Kelly et al., 2010) has the goal to
build a database, a wiki, and an ontology of rhetorical
figures. Based on this work, an ontology for rhetorical
figures in the Serbian language was modelled in the
Web Ontology Language (OWL) (McGuinness et al.,
2004). This ontology is called RetFig and aims to
model all rhetorical figures in the Serbian language
(Mladenovi
´
c and Mitrovi
´
c, 2013). Another ontology
was developed to spot rhetorical figures of speech in
text (O’Reilly and Paurobally, 2010). The ontology
was modelled using OWL and Semantic Web Rule
Language (SWRL) rules (Horrocks et al., 2004). A
comparison of different forms and rhetorical concepts
shows the context between the Rhetorical Structure
Theory (RST), the Serbian RetFig ontology, and the
Lassoing Rhetoric project (Mitrovi
´
c et al., 2017). An
ontology for the rhetorical figure litotes was modelled
by (Mitrovic et al., 2020).
A top-down, middle-out, and a bottom-up ap-
proach to model an ontology under neuro-cognitive
aspects of rhetorical figures is presented by (Harris
et al., 2017). As future work, the authors would like
to connect their ontology with others, especially “lin-
guistic and cognitive ontologies”. Based on this work,
an OWL ontology for argumentation was developed
for a suite of rhetorical figures (O’Reilly et al., 2018;
Black et al., 2019). The intersection of multiple fig-
ures is considered. The ontology was developed for
a compound rhetorical figure called climax, in which
words or word groups are repeated with a semantic
increase, e.g., “Minutes are hours there, and the hours
are days, / Each day’s a year, and every year an age”
(Suckling, Aglaura, 3.2) (Suckling, 1637). It con-
sists of a trope called incrementum, in which words
or word groups form a semantic increase in a pas-
sage. The figure climax also consists of two rhetorical
figures of perfect lexical repetition, i.e., gradatio and
anadiplosis. An anadiplosis is defined as “the repe-
tition of a word or word sequence on both sides of
a clause or phrase boundary” (Harris and Di Marco,
2017), e.g., “I beg your pardon. Pardon, I beseech
you!” (Shakespeare, Romeo and Juliet, 4.2) (Shake-
speare, 2014). A gradatio features a sequence of
anadiploses (Harris and Di Marco, 2017). The au-
thors defined anadiplosis and gradatio in terms of el-
ements, colons, and tokens. This model was then
extended to construct the Ploke ontology that cov-
ers other types of rhetorical figures of perfect lexical
repetition and incorporates notions of neuro-cognitive
affinities such as repetition and position (Wang et al.,
2021). By incorporating views from (Harris, 2020),
(Wang et al., 2021) demonstrated that it is possible to
model how simple syntactical patterns such as perfect
lexical repetition can generate attentional effects such
as salience, aesthetic pleasure, and mnemonic effect.
This paper focuses on the future work mentioned
in several papers (Harris et al., 2017; Mladenovi
´
c and
Mitrovi
´
c, 2013; Wang et al., 2021): Combining dif-
ferent rhetorical figure ontologies. We combine the
Ploke ontology with the Serbian RetFig ontology and
its adaption to the German GRhOOT ontology. The
differences between the models will be highlighted
as well as the peculiarities that arise when combining
them into a multilingual ontology. Note that this pa-
per does not focus on ontology matching and merging
techniques since the ontologies and corresponding in-
consistencies involved are particular to the domain of
rhetorical figures. However, procedures discussed in
this paper are inevitably applications of general on-
tology matching and merging techniques. Readers
may refer to (Euzenat and Shvaiko, 2013) for a com-
prehensive introduction to general ontology matching
and merging methods.
3 COMPARISON OF RetFig,
GRhOOT, AND PLOKE
ONTOLOGY
In this section, we discuss the conceptual models of
the RetFig, GRhOOT, and Ploke ontology. Further,
we provide a comparison of rhetorical figure defi-
nitions in the RetFig (Serbian), GRhOOT (German)
and Ploke ontologies. In this paper, we use capital-
ized words to indicate a class name (e.g., Class), and
typewriter lower case words to indicate an individual
name (e.g., ind1) in an ontology. Names of rhetorical
figures are italicized on the first appearance.
3.1 Concepts of RetFig/GRhOOT:
Serbian and German Ontologies
RetFig (Mladenovi
´
c and Mitrovi
´
c, 2013) is a formal
domain ontology for rhetorical figures in the Serbian
language, written in OWL. Figures are classified ac-
cording to rhetorical and linguistic types and rhetori-
cal operations. It was inspired by the ontology devel-
oped by (Harris and DiMarco, 2009) and (Kelly et al.,
2010) in the scope of the RhetFig project. There,
rhetorical figures are classified based on Linguistic
Domain (e.g., phonological, morphological, and lex-
ical), neuro-cognitive pattern biases (e.g., repetition,
position, and similarity), and traditional categories
(e.g., trope and scheme). A figure can belong to more
than one class.
In the Serbian RetFig, the authors reduced the
number of Linguistic Entities to four: phonological,
Towards a Unified Multilingual Ontology for Rhetorical Figures
119

Table 1: Explanation of figures in different languages.
Figure Language Definition
Anadiplosis English The repetition of a word or word group on both sides of a clause or phrase
boundary.
Anadiplose German Repetition of the last word of a sentence at the beginning of the next
sentence.
Palilogija Serbian Repetition of the words at the end of a verse at the beginning of the next
verse.
Conduplicatio English Unpatterned perfect lexical repetition.
Epanalepse German Repetition of a word or group of words in a sentence (Dudenredaktion,
2022).
Epanalepsis English The repetition of a word or word group at the beginning and ending of
the same clause or phrase.
Epanadiplose German Repetition of a word at the beginning and end of a sentence/line.
Okruzivanje Serbian The same word is at the beginning and end of the sentence (verse).
Epanaphora English The occurrence of the same word or word group at the beginning of
proximal clauses or phrases.
Anapher German Repetition of the word in the beginning in successive sentences.
Anafora Serbian Repetition of the same words in the beginning (verse).
Epiphora English The occurrence of the same word or word group at the end of proximal
clauses or phrases.
Epipher German Inversed anaphora.
Epifora Serbian Repetition of the same words at the end (verses).
Epizeuxis English The immediate repetition of a word or word group with no other words
intervening.
Epizeuxis German Three or more repetitions of the same word/word group.
Epizeusa Serbian Repeating the same word several times in the same sentence or verse for
emphasis.
Gradatio English A sequence of anadiploses.
Gradation German Gradual increase, top term for climax/anticlimax.
Gradacija Serbian Transferring one word or word group from one sentence or verse to the
next sentence or verse in order to gradually increase or decrease the
strength of the initial statement.
Mesodiplosis English Lexical repetition of a word or word group in the middle of clauses or
phrases.
Symploce/Symploke English Combination of epanaphora and epiphora: The occurrence of the same
word or word group at the beginnings of two or more proximal phrases
or clauses while another word or word group occurs at the ends of those
same phrases or clauses.
Symploke German Combination of anaphora and epiphora.
Simploka Serbian Repetition of the same words at the beginning and at the end (verses).
Anaklaza Serbian Repeating the same word in the same verse for emphasis.
Antiklimax German Downward gradatio.
Emphase German Emphatic emphasis of a word, often acoustically or with exclamation
marks.
Emfaza Serbian Emphasis on words in a sentence. Exaggeration in the tone or expression
with which the writer or speaker emphasizes certain words, thoughts or
feelings. It can turn into force.
KEOD 2022 - 14th International Conference on Knowledge Engineering and Ontology Development
120

morphological, pragmatic, and syntactic, and
do not concern themselves with neuro-cognition or
traditional categories. Instead, they modelled rhetori-
cal operations. The overall structure is shown in Fig-
ure 1. Each figure is assigned a LinguisticEntity and
a RhetoricalEntity. A LinguisticEntity has multiple
subclasses such as LinguisticObject, LinguisticScope,
LinguisticElement, etc. A rhetorical operation desc-
ribes how a rhetorical figure is formed, e.g.,
addition, omission, repetition, transposition,
joining, separation, or symmetry. These oper-
ations are similar to the Kind-Of classification in
RhetFig. For the RhetoricalEntity, RetFig differenti-
ates between figures of pronunciation, of meaning
tropes, of construction, and of thoughts. For
each figure, an English name is provided such that a
mapping to other linguistic ontologies is possible in
the future.
GRhOOT, the German RhetOrical OnTology, is
the adaption of the Serbian RetFig to the German
language. GRhOOT uses the same structure as Ret-
Fig (cf. Figure 1), however, some figures have been
adapted/added/omitted according to differences in the
definitions of figures in the Serbian and the German
language.
3.2 Concepts of the Ploke Ontology
The Ploke ontology (Wang et al., 2021), as the
name suggests, deals with ploke, rhetorical figures
of perfect lexical repetition. At the top level, it
defines a class RhetoricalFigure, Form, and Neuro-
cognitiveAffinity with relations to depict a statement
which we refer to as the “neuro-cognitive path”: a
rhetorical figure has a form, the form triggers a set
of neuro-cognitive affinities that generate a set of at-
tentional and mnemonic effects (Wang et al., 2021).
Neuro-cognitive affinities are patterns that can be eas-
ily recognized by a human mind, generating atten-
tional and mnemonic effects that make certain con-
cepts more salient, aesthetically pleasing, and more
memorable (Harris et al., 2017). This is illustrated
by the highlighted path in Figure 2. The Ploke ontol-
ogy focuses on representing concepts specific to ploke
along this path.
Instead of treating ploke as a single figure, the
Ploke ontology adopts the view of Harris in which
ploke should be treated as a class of figures of perfect
lexical repetition (Harris, 2020). Figure 2 outlines
classes and relations in the Ploke ontology. A class
Ploke is defined as a subclass of RhetoricalFigure
and has a number of subclasses to represent the fig-
ures of perfect lexical repetition such as epanaphora,
epizeuxis, mesodiplosis, antimetabole, etc.
2
They
are defined in terms of their positions relative to the
clause structure, e.g., an epanaphora is a figure in
which words or word groups are repeated at the be-
ginning of a clause; or respective to other lexical
items, e.g., epizeuxis is a figure in which a word or
word group is repeated immediately. Each class as-
sociates with a corresponding subclass of Form. Two
classes, Repetition and Position are defined as sub-
classes of Neuro-cognitiveAffinity, each of which has
subclasses representing different types of lexical rep-
etition corresponding to subclasses of Ploke as shown
in Figure 4. There are other neuro-cognitive affini-
ties but repetition and position are the two most fun-
damental affinities triggered by ploke (Harris, 2020;
Wang et al., 2021). For example, in the Ploke ontol-
ogy, a class Epizeuxis is connected to an Epizeuxis-
Form, which triggers ImmediateRepetition. Each of
these is a subclass of Ploke, PlokeForm, and Rep-
etition respectively, which in turn are subclasses of
RhetoricalFigure, Form, and Neuro-cognitiveAffinity,
respectively. The hierarchy is illustrated in Figure 3.
3
As an enhancement of the Ploke ontology, a Repeti-
tion can be further classified into classes SingleRepe-
tition or NestedRepetition. A SingleRepetition is con-
nected to a class Element that refer to a word or word
group and has a position. A NestedRepetition is a
subclass of Repetition in which the elements are also
isntances of Repetition. For simplicity, we omit Nest-
edRepetition and refer to SingleRepetition as Repeti-
tion in this paper.
3.3 Figures of Perfect Lexical
Repetition in the Ontologies
For each figure of perfect lexical repetition, we com-
pared within the three ontologies if it exists in the
other languages in general and if it is modelled in the
ontologies, too. The result is shown in Table 2. A
dash indicates that a figure does not exist in this on-
tology.
Some figures like anaklaza, mesodiplosis, or an-
tiklimax, are only modelled in one ontology or only
in one of the ontologies considered as ploke. Only in
GRhOOT, antiklimax is considered as a figure of per-
fect lexical repetition, but klimax (climax) is not. It is
questionable if this is just due to a wrong classifica-
tion. One might notice, for instance, that epanalepsis
is not the same as the German figure epanalepse, but
epanadiplose. The terminology of rhetorical figures
is notoriously inconsistent. We want to look further
2
Definitions are listed in Table 1.
3
Relation names are omitted for readability.
Towards a Unified Multilingual Ontology for Rhetorical Figures
121

Figure 1: Structure of the RetFig ontology, adapted from (Mladenovi
´
c and Mitrovi
´
c, 2013).
Ploke
Anadiplosis Mesodiplosis Antimetabole ...
Ploke Form
Repetition
Conduplicatio
Rhetorical
Figure
Form
Neuro-
cognitive
Affinity
Position
Attentional
Effect
sublcass of
class
binary relation
Figure 2: Classes related to ploke in the Ploke ontology,
adapted from (Wang et al., 2021).
Ploke
Epizeuxis
Epizeuxis
Form
Repetition
Immediate
Repetition
Ploke
Form
Rhetorical
Figure
Neuro-cognitive
Affinity
Form
Element
sublcass of
class
binary relation
Figure 3: Classes related to Epizeuxis in the Ploke ontology.
Table 2: Comparison of figures of perfect lexical repetition.
Ploke Ontology Serbian RetFig German GRhOOT
Anadiplosis Palilogija Anadiplose
Conduplicatio – Epanalepse
Epanalepsis Okruzivanje Epanadiplose
Epanaphora Anafora Anapher
Epiphora Epifora Epipher
Epizeuxis Epizeusa Epizeuxis
Gradatio Gradacija Gradation
Mesodiplosis – –
Symploke Simploka Symploke
– Anaklaza –
– – Antiklimax
– Emfaza Emphase
into the definitions to spot differences and detect in-
consistencies. Table 1 provides the definitions from
each ontology for the respective figure. We provide
the English definitions (Harris and Di Marco, 2017)
that are used in the Ploke ontology, the German def-
initions (Berner, 2011) from GRhOOT, and the Ser-
bian definitions from the RetFig ontology.
4 INCONSISTENCIES OF THE
ONTOLOGIES
This section discusses types of inconsistencies be-
tween RetFig/GRhOOT and the Ploke ontology.
These inconsistencies fall under a more general clas-
sification of heterogeneity outlined by (Euzenat and
Shvaiko, 2013). We adapt those types of heterogene-
ity to introduce the following types of inconsistencies.
4.1 Model Inconsistency
Model inconsistency is a type of conceptual or seman-
tic heterogeneity, particularly an instance of differ-
ence in granularity or difference in perspective (Eu-
zenat and Shvaiko, 2013). The difference is caused
by the level of details represented for concepts in
each ontology. The most significant inconsistency
is that rhetorical figures are modelled as individu-
als in GRhOOT, but are modelled as classes in the
Ploke ontology. The former focuses on the relation
of rhetorical figures with linguistic entities in gen-
eral, in which it is sufficient to represent each rhetor-
ical figure as an individual. The latter models a spe-
cific class of figures, i.e., ploke, which requires the
representation of the hierarchical relations between
ploke and its subclasses, and their relations to cor-
responding subclasses of repetition and position. For
example, in GRhOOT, the figure epizeuxis is repre-
sented by an individual epizeuxis, an instance of the
class RhetoricalFigure, with axioms shown in Table 3.
In the Ploke ontology, the figure epizeuxis is repre-
sented by a class Epizeuxis, an indirect subclass of
the class RhetoricalFigure. An instance of Epizeuxis
is an individual representing a specific epizeuxis,
e.g., epiz1, which is connected to epiz1form and
immediateRep1 which are instances of the Epizeux-
isForm and ImmediateReptition, respectively. The re-
lations are outlined in Figure 5.
In order to resolve this inconsistency, classes that
are defined in the Ploke ontology will replace the
corresponding individuals in GRhOOT while main-
KEOD 2022 - 14th International Conference on Knowledge Engineering and Ontology Development
122

Single
Repetition
Neuro-cognitive
Affinity
Relative
Repetition
Nested
Repetition
Cross-Boundary
Repetition
Constituent-
Medial Repetition
Constituent-
Final Repetition
Immediate
Repetition
Reversed
Repetition
Gradatio
Repetition
Element
Word/Word
Group
refers
To
Unpatterned
Repetition
Respective
Repetition
Between
Repetition
Constituent-
Initial Repetition
Outer-Boundary
Repetition
Position
sublcass of
class
binary relation
Repetition
Figure 4: Repetition in the Ploke ontology, modified based on (Wang et al., 2021).
taining conceptual and ontological inconsistencies as
discussed in the following sections. Individuals of
rhetorical figures that are not part of the Ploke on-
tology remain untouched. These individuals can be
used as placeholders for future extension of potential
ontologies in the rhetorical figure domain.
4.2 Conceptual Inconsistency
Conceptual inconsistency is a combination of concep-
tual heterogeneity and terminological heterogeneity
(Euzenat and Shvaiko, 2013). It is caused by dif-
ferences in definitions of rhetorical figures and the
scopes of the ontologies. Conceptual inconsistency
can be terminological or due to differences in cover-
age.
4.2.1 Terminological Inconsistency
Terminological inconsistencies occur when different
names are used for the same rhetorical figures, or
when a rhetorical figure is represented as different en-
tities. For example, epizeuxis is defined as a figure
in which words or word groups are repeated imme-
diately in the Ploke ontology, but defined as a figure
in which words or word groups are repeated three or
more times in GRhOOT.
In order to resolve terminological inconsistencies,
the definitions will be carefully compared. In the case
of epizeuxis as mentioned above, we first distinguish
representations of epizeuxis in the ontologies by cre-
ating a class EpizeuxisG for epizeuxis in GRhOOT.
The class Epizeuxis follows the definition in Figure 3
in the Ploke ontology.
Figure 5 demonstrates how epiz1, an instance of
Epizeuxis, is represented graphically using the phrase
“long, long ago”. Recall that epiz1 is connected
to immediateRep1, an instance of ImmediateRepeti-
tion, which is connected to two instances of Element,
Immediate
Repetition
immediateRep1
e1 e2
position(e1) position(e2)
Position
refers
To
"long"
epiz1
epiz1form
Epizeuxis
Form
Epizeuxis
binary relation
class
instance of
individual
Immediate
Repetition
immediateRep1
e1
position(e1)
"long"
Position
position(e2)
e2
epiz1
immediately precedes
Figure 5: An instance of Epizeuxis in the Ploke ontology of
the phrase “long, long ago”.
e1 and e2. The Element class represents each occur-
rence of the repeated word or word group. An in-
stance of Element refers to an instance of the class
Word/WordGroup and is connected to an instance of
the class Position. The position of the first element
of “long”, i.e., e1, immediately precedes the second
element, i.e., e2. Both e1 and e2 refer to the word
“long”. Implicitly, the position of e2 immediately
follows the position of e1. There are two cases we
consider for the definition of EpizeuxisG:
1. Words or word groups repeat 3 or more times.
2. Words or word groups repeat 3 or more times. im-
mediately, e.g., “a long, long, long time ago”.
In case 1, EpizeuxisG is not necessarily an
epizeuxis, it is a subclass of Ploke distinguished from
Epizeuxis. A new class Ploke 3 is created as a sub-
class of Ploke with additional cardinality constraints
on number of occurrences associated with the corre-
sponding subclass of Repetition. In case 2, Epizeux-
isG is a more specified version of both the class
Ploke 3 and Epizeuxis, therefore a subclass of both.
Towards a Unified Multilingual Ontology for Rhetorical Figures
123

The resulting classes and relations are shown in Fig-
ure 6.
Another example of terminological inconsistency
regards the figure epanalepsis. Its corresponding Ser-
bian term is okruzivanje, but in German the name
is epanadiplose, whereas an epanalepse in German
is just a repetition of words without any constraints,
similar to conduplicatio in English. We create
classes EpanadiploseG and EpanalepseG to represent
epanadiplose and epanalepse defined in GRhOOT re-
spectively. Conduplicatio and Epanalepsis are already
subclasses of the class Ploke in the Ploke Ontology
as shown in Figure 2. Therefore, we construct the
following mappings from GRhOOT to the Ploke on-
tology regarding epanalepsis: 1) EpanadiploseG →
Epanalepsis; 2) EpanalepseG → Conduplicatio.
4.2.2 Inconsistency in Coverage
Different sets of rhetorical figures are included in each
ontology due to difference in coverage and perspec-
tive (Euzenat and Shvaiko, 2013). In order to resolve
coverage inconsistencies, each disjoint figure listed in
Section 3.3 will be checked individually. Since the
Ploke ontology has the most specific details, it is used
as baseline ontology. We check whether the figures
from GRhOOT can fit into the model of the Ploke on-
tology. For example, since mesodiplosis is the only
figure that is only present in the Ploke ontology and
as this ontology serves as baseline, thus no further ac-
tions need to be taken for mesodiplosis.
On the other hand, there are three figures included
in the Serbian RetFig and GRhOOT but not in the
Ploke ontology, i.e., antiklimax, anaklaza, and em-
phase. Antiklimax, an inversed version of the figure
climax, is not considered as a figure of perfect lexical
repetition in the Ploke ontology. Recall from Section
2, climax is a figure that consists of anadiplosis and
gradatio but also incrementum which is not a figure of
perfect lexical repetition (O’Reilly et al., 2018; Black
et al., 2019). Therefore, the Ploke ontology includes
only classes Anadiplosis and Gradatio as subclasses
of Ploke, but no class to represent incrementum nor
climax. Anaklaza and emphase are figures of perfect
lexical repetition that are not represented in the Ploke
ontology which will be discussed in the following sec-
tion.
4.3 Ontological Inconsistency
This type of inconsistency deals with domain-specific
concepts that are represented differently in each on-
tology due to the different purposes and scopes when
modelling the ontologies. This is referred to as
semiotic or pragmatic heterogeneity (Euzenat and
Shvaiko, 2013). A significant ontological inconsis-
tency is how repetition is represented in each ontol-
ogy. In the Serbian RetFig and GRhOOT ontologies,
repetition is merely implicitly defined as object prop-
erties. Four different subproperties of the repetition
property are differentiated here: Repetition of another
form, of the same form, with different meaning, of
different kind. Individual elements like word, sen-
tence, or phrase can be assigned to those properties.
In the Ploke ontology, repetition is represented as a
subclass of Neuro-cognitiveAffinity as demonstrated
in Figure 5. It includes more details such as sub-
classes to represent different types of repetition, the
class Element to represent occurrences of repeated
words or word groups, and their positions. There-
fore, the ontologies are merged based on the Ploke on-
tology while incorporating any suitable axioms from
GRhOOT regarding a specific rhetorical figure of per-
fect lexical repetition. Again, we demonstrate this
with epizeuxis as an example. In GRhOOT, an in-
dividual Epizeuxis is defined with the axioms in Ta-
ble 3, where all values are individuals. Repetition and
position orientation of epizeuxis are only implicitly
defined with the object properties isARepeatableEle-
mentOfTheSameForm and isInPosition.
Table 3: Axioms of Epizeuxis in GRhOOT.
Object Property Value
isARepeatableElement
OfTheSameForm
Wordelement
isInObject Wordobject
isInPosition Whole-Succesive
isLinguisticGroup Syntactic
isRhetoricalGroup FigureOfSpeech
isInArea Sentence
isInArea Verse
As shown in Figure 5, the Ploke ontology de-
fines repetition with an explicit and more detailed
definition such as each occurrence of the repeated
words or word groups. A class ImmediateRepeti-
tion, a subclass of Repetition, is used to represent
the specific type of repetition triggered by the form of
an epizeuxis represented by an EpizeuxisForm class.
The positions of the repeated words or word groups
are also defined with a Position class. An object prop-
erty immediatelyPrecedes, which is a subproperty of
precedes (similarly for immediatelyFollows and fol-
lows) represents the orientation of positions explicitly.
Another example of ontological inconsistency is
the rhetorical figure emphase in GRhOOT. It is de-
fined as emphasis of a word, often acoustically or
with exclamation mark (Berner, 2011). It deals with
phonological or orthographic methods of applying an
emphasis effect. In the case of ploke, the emphasis
KEOD 2022 - 14th International Conference on Knowledge Engineering and Ontology Development
124

Epizeuxis
Repetition_3
Element
3,*
Ploke
Ploke_3
Ploke_3
Form
Single
Repetition
Ploke
Form
EpizeuxisG
2,*
sublcass of
class
binary relation
Repetition
Figure 6: New classes Ploke 3 and EpizeuxisG added to resolve Case 1 and 2 respectively.
effect naturally applies to every ploke since the re-
peated words or word groups would already increase,
to some extent, the salience of the concept carried as
part of the attentional effect (Harris, 2020), thus em-
phasizing the concept of the repeated word or word
group. The Ploke ontology implicitly represents this
emphasis effect as part of the neuro-cognitive path
(the highlighted path in Figure 2). That is, the form
of a rhetorical figure triggers neuro-cognitive affini-
ties (e.g., repetition) which generate attentional and
mnemonic effect (e.g., salience). Since the Ploke on-
tology does not deal with phonological, orthographic,
or punctuation aspects of natural language, it does not
include emphase as part of its representation. This
also suggests that the figure anaklaza is equivalent
to conduplicatio under this representation, since the
only difference between the definitions is the notion
of ‘emphasis’ as shown in Table 1.
Neuro-cognitive
Affinity
Repetition Position
......Metaphor Rhyme Sarcasm
Linguistic EntitiesRhetorical Entities
Ploke
Epanaphora Anadiplosis
......
Epizeuxis
Antimetabole
RetFig/
GRhOOT
+
Ploke
RetFig/
GRhOOT
Ploke
Figure 7: Combination of the RetFig, GRhOOT, and Ploke
ontologies.
5 COMBINING THE RetFig,
GRhOOT, AND PLOKE
ONTOLOGIES
The resulting ontology is a combination of the Ser-
bian RetFig, GRhOOT and Ploke ontology as shown
Figure 8: Prot
´
eg
´
e’s entity graph for the figure Epizeuxis.
in Figure 7. Concepts that apply to only one ontol-
ogy remain unchanged, e.g., subclasses of Linguis-
ticEntities and Neuro-cognitiveAffinity, respectively.
The middle section of Figure 7 indicates the over-
lapping concepts in all three ontologies. These con-
cepts are subclasses of ploke represented by newly
formed classes to incorporate existing concepts from
both GRhOOT and the Ploke ontology. Note that only
subclassOf relations are shown in Figure 7. The OWL
ontology was implemented using WebProt
´
eg
´
e (Tudo-
rache et al., 2013). Figure 8 shows the entity graph
of the figure epizeuxis, represented by an individual
Epizeuxis with a type of the class Epizeuxis. The
axioms from GRhOOT shown in Table 3 are present
here again. Additionally, the individual represent-
ing epizeuxis is now connected to a class of Rhetor-
icalFigure whose form triggers ImmediateRepetition.
This is a subclass of Repetition, which in turn is a
subclass of Neuro-cognitiveAffinity. As Epizeuxis is
a type of Ploke, it is marked as a subclass of Ploke
which is a subclass of the rhetorical figures of the cat-
egory Scheme. These are all rhetorical entities be-
longing to the top class RhetoricalEntity.
Towards a Unified Multilingual Ontology for Rhetorical Figures
125

6 CONCLUSION
The resulting ontology for rhetorical figures of perfect
lexical repetition is a combination of three different
ontologies that tackles not only multilingual differ-
ences but also conceptual and terminological incon-
sistencies. It is tightly connected but still modular.
We have shown the representations of different fig-
ures of perfect lexical repetition in the ontologies and
their definitions. The inconsistencies were identified
to be later resolved.
Future work includes the establishment of a gen-
eralized framework for ontologies in the domain of
rhetorical figures, as also suggested by (Mitrovi
´
c
et al., 2017). This framework could be similar to the
CIDOC-Conceptual Reference Model (Doerr, 2003)
that is used in the domain of cultural heritage infor-
mation. It is an ontology which is restricted to prede-
fined semantics that are specific in the domain of cul-
tural heritage. For the domain of rhetorical figures, a
standard notation of text elements is required (Harris
and DiMarco, 2009). Furthermore, more figures can
be combined in one unified ontology.
We paved the way for combining ontologies of
rhetorical figures in a potentially generalizable way:
It involves much domain-specific knowledge. We
hope for an automated tool that incorporates rhetori-
cal figure domain knowledge as more rhetorical figure
ontologies are being developed.
ACKNOWLEDGMENTS
The project on which this report is
partly based was funded by the So-
cial Sciences and Humanites Research
Council of Canada and the German
Federal Ministry of Education and Research (BMBF)
under the funding code 01—S20049. The authors are
responsible for the content of this publication.
REFERENCES
Alliheedi, M. and Di Marco, C. (2014). Rhetorical figu-
ration as a metric in text summarization. In Cana-
dian Conference on Artificial Intelligence, pages 13–
22. Springer.
Berner, G. (2011). Vollst
¨
andiges Kompendium der
rhetorischen Mittel, Stilfiguren und Tropen f
¨
ur Ober-
stufensch
¨
uler und Studienanf
¨
anger. GRIN Verlag,
M
¨
unchen.
Black, L. A., Tu, K., O’Reilly, C., Wang, Y., Pacheco, P.,
and Harris, R. A. (2019). An ontological approach to
meaning making through path and gestalt foreground-
ing in climax. The American Journal of Semiotics,
35(1/2):217–249.
Caselli, T., Basile, V., Mitrovi
´
c, J., Kartoziya, I., and
Granitzer, M. (2020). I feel offended, don’t be abu-
sive! implicit/explicit messages in offensive and abu-
sive language. In Proceedings of the 12th Lan-
guage Resources and Evaluation Conference, pages
6193–6202, Marseille, France. European Language
Resources Association.
Clarke, E. L. (2019). The Relo-KT Process for Cross-
Disciplinary Knowledge Transfer. PhD thesis, Trinity
College Dublin, Ireland.
Doerr, M. (2003). The cidoc conceptual reference module:
an ontological approach to semantic interoperability
of metadata. AI magazine, 24(3):75–75.
Dudenredaktion (2022). Epanalepse. https://www.duden.
de/rechtschreibung/Epanalepse. Accessed: 2022-02-
17.
Euzenat, J. and Shvaiko, P. (2013). Ontology Matching.
Springer Science & Business Media.
Fahnestock, J. (2002). Rhetorical Figures in Science. Ox-
ford University Press on Demand.
Green, N. L. (2020). Recognizing rhetoric in science pol-
icy arguments. Argument & Computation, 11(3):257–
268.
Green, N. L. and Crotts, L. J. (2020). Towards automatic
detection of antithesis. In Computational Models of
Natural Argument, pages 69–73.
Harris, R. and DiMarco, C. (2009). Constructing a rhetori-
cal figuration ontology. In Persuasive Technology and
Digital Behaviour Intervention Symposium, pages 47–
52.
Harris, R. A. (2020). Ploke. Metaphor and Symbol,
35(1):23–42.
Harris, R. A. and Di Marco, C. (2017). Rhetorical figures,
arguments, computation. Argument & Computation,
8(3):211–231.
Harris, R. A., Di Marco, C., Mehlenbacher, A. R., Clap-
perton, R., Choi, I., Li, I., Ruan, S., and O’Reilly,
C. (2017). A cognitive ontology of rhetorical figures.
Cognition and Ontologies, pages 18–21.
Horrocks, I., Patel-Schneider, P. F., Boley, H., Tabet, S.,
Grosof, B., Dean, M., et al. (2004). Swrl: A semantic
web rule language combining owl and ruleml. W3C
Member submission, 21(79):1–31.
Java, J. (2015). Characterization of Prose by Rhetorical
Structure for Machine Learning Classification. PhD
thesis, Nova Southeastern University.
Karp, M., Kunanets, N., and Kucher, Y. (2021). Meiosis
and litotes in the catcher in the rye by jerome david
salinger: text mining. In CEUR Workshop Proceed-
ings, volume 2870, pages 166–178.
Kelly, A. R., Abbott, N. A., Harris, R. A., DiMarco, C.,
and Cheriton, D. R. (2010). Toward an ontology of
rhetorical figures. In Proceedings of the 28th ACM In-
ternational Conference on Design of Communication,
pages 123–130.
K
¨
uhn, R., Mitrovi
´
c, J., and Granitzer, M. (2022). GRhOOT:
Ontology of Rhetorical Figures in German. In Pro-
ceedings of The 13th Language Resources and Eval-
KEOD 2022 - 14th International Conference on Knowledge Engineering and Ontology Development
126

uation Conference, Marseille, France. European Lan-
guage Resources Association.
Lagutina, N. S., Lagutina, K. V., Boychuk, E. I.,
Vorontsova, I. A., and Paramonov, I. V. (2020). Auto-
mated rhythmic device search in literary texts applied
to comparing original and translated texts as exempli-
fied by english to russian translations. Automatic Con-
trol and Computer Sciences, 54(7):697–711.
Lawrence, J., Visser, J., and Reed, C. (2017). Harnessing
rhetorical figures for argument mining. Argument &
Computation, 8(3):289–310.
Lemmens, J., Markov, I., and Daelemans, W. (2021). Im-
proving hate speech type and target detection with
hateful metaphor features. In Proceedings of the
Fourth Workshop on NLP for Internet Freedom: Cen-
sorship, Disinformation, and Propaganda, pages 7–
16.
McGuinness, D. L., Van Harmelen, F., et al. (2004). Owl
web ontology language overview. W3C recommenda-
tion, 10(10):2004.
Mitrovic, J., O’Reilly, C., Harris, R. A., and Granitzer, M.
(2020). Cognitive modeling in computational rhetoric:
Litotes, containment and the unexcluded middle. In
ICAART (2), pages 806–813.
Mitrovi
´
c, J., O’Reilly, C., Mladenovi
´
c, M., and Handschuh,
S. (2017). Ontological representations of rhetorical
figures for argument mining. Argument & Computa-
tion, 8(3):267–287.
Mladenovi
´
c, M. and Mitrovi
´
c, J. (2013). Ontology of
rhetorical figures for serbian. In International Confer-
ence on Text, Speech and Dialogue, pages 386–393.
Springer.
Musolff, A. (2015). Dehumanizing metaphors in uk immi-
grant debates in press and online media. Journal of
Language Aggression and Conflict, 3(1):41–56.
O’Reilly, C. and Paurobally, S. (2010). Lassoing rhetoric
with owl and swrl. Unpublished MSc dissertation.
Available: http://computationalrhetoricworkshop.
uwaterloo.ca/wpcontent/uploads/2016/06/
LassoingRhetoricWithOWLAndSWRL.pdf .
O’Reilly, C., Wang, Y., Tu, K., Bott, S., Pacheco, P., Black,
T. W., and Harris, R. A. (2018). Arguments in grada-
tio, incrementum and climax; a climax ontology. In
Proceedings of the 18th workshop on Computational
Models of Natural Argument. Academic Press.
Shakespeare, W. (2014). The complete works of William
Shakespeare. Race Point Publishing.
Strommer, C. W. (2011). Using Rhetorical Figures and
Shallow Attributes as a Metric of Intent in Text. PhD
thesis, University of Waterloo.
Suckling, S. J. (c. 1637). Aglaura.
Tudorache, T., Nyulas, C., Noy, N. F., and Musen, M. A.
(2013). Webprot
´
eg
´
e: A collaborative ontology editor
and knowledge acquisition tool for the web. Semantic
web, 4(1):89–99.
Wang, Y., Harris, R. A., and Berry, D. M. (2021). An ontol-
ogy for ploke: Rhetorical figures of lexical repetitions.
CAOS 2021: 5th Workshop on Cognition And Ontolo-
gieS.
Towards a Unified Multilingual Ontology for Rhetorical Figures
127