POLITICAL RHETORIC IN ELECTIONS

Abstract

This study describes a tool for identifying the prevailing rhetorical tones in electoral speeches (here the 2014 presidential campaign in Romania), especially from the print press. The application, abbreviated PDA, allows for a rapid and robust interpretation of the electoral language, requiring an interdisciplinary approach. By emphasizing the rhetoric component at the level of speech, electors identify with the candidate, who becomes the personification of their common expectations. Rhetoric diversity is an important problem for receiving message, due to the heterogeneity of auditors. This investigation is intended to give support to specialists in political sciences, to political analysts, sociologists and election's staff, being helpful mainly in their interpretation of the electoral campaigns, as well as to the media, in their intend to evaluate reactions with respect to the developments in the political scene.

Full text

POLITICAL RHETORIC IN ELECTIONS
Daniela Gîfu
Faculty of Computer Science, “Alexandru Ioan Cuza” University of Iași, Romania
Center for Advanced Research in Applied Informatics, University of Craiova, Romania
Abstract. This study describes a tool for identifying the prevailing
rhetorical tones in electoral speeches (here the 2014 presidential
campaign in Romania), especially from the print press. The application,
abbreviated PDA, allows for a rapid and robust interpretation of the
electoral language, requiring an interdisciplinary approach. By
emphasizing the rhetoric component at the level of speech, electors
identify with the candidate, who becomes the personification of their
common expectations. Rhetoric diversity is an important problem for
receiving message, due to the heterogeneity of auditors. This
investigation is intended to give support to specialists in political
sciences, to political analysts, sociologists and election’s staff, being
helpful mainly in their interpretation of the electoral campaigns, as well
as to the media, in their intend to evaluate reactions with respect to the
developments in the political scene.
Keywords: discursive rhetoric, semantic classes, resources, elections, print press
1 Introduction
The motivation for this study relies on the need for objectivity in the
interpretation of the election speech, situated at the intersection of the
political space, the public space and the communicational space (Wolton,
1998), as well as on the need to measure to what extent and in what ways
a speech can influence its direct receptor. In other words, this study
highlights the relevance and understanding forms of communication,
captured by the print press in the electoral context. To interpret correctly
an electoral phenomenon means to understand how the rhetoric works.
This kind of approach moves from persuasion to aesthetic, being “the art
of invention, choice and ornaments convenient expression that can serve
to convince” (Quintilian, 1971).
First of all we must take into account the past events. Each electoral
event "is an action that always tends to alter a preexisting condition"
(Perelman and Tyteca, 1970: 72).

2
Second, we can consider an election speech like an "aggressor",
because it promotes and supports the programs and values of a group of
power, capable to answer the elector’s interests. Electors demand from
politicians the logicians’ open mind, the philosophers’ deep meditation,
the poets’ metaphoric expression, the jurists’ bright memory, tragedians’
penetrant voice and, I’d say, a famous actor’s gestures. (Cicero, 1973:
51).
Third, the basic assumption in the discourse analysis is that any text
isn't merely a string of signs placed randomly. Any group of signs is
hierarchically organized, the signs can define various informational and
interaction relations (Fox, 1987).
Considering the Habermas’ the communication the public domain
comes increasingly under the control of private business interests, either
through direct and interactive forms, such as phone or Internet, or by
means of mass communication, centrally controlled, such as audiovisual
and print press. The deviation in terms of rules of construction may be, on
one hand, deliberate, so as to achieve specific rhetorical or aesthetic
purposes, or, on the other hand, be an expression of social and cognitive
characteristics of those who use language such as “memory limitations,
the strategic aspects of speech production” (van Dijk, 1972: 14).
Currently, NLP (Natural Language Processing) proves the
applicability of technologies created by IT specialists in the sphere of
social and human sciences (SSU) of the utmost importance (Cristea,
Tufiş, 2002: 211-234). As in this case, dealing with word-processing,
SSU benefits that researchers can get use of these technologies in
language interpretation and understanding of the discursive context.
Among many attributes the political discourse has in print media (i.e.
the editorials), we were interested in the lexicon and its interpretation in a
range of semantic coordinates. The final objective of this study is to
develop a computational framework able to offer to the researchers in
mass-media, political sciences, political analysts, to the public at large
(interested to consolidate their options before elections), and, why not,
even to politicians themselves, the possibility to measure different
parameters of a written political discourse. Part of this research, as
reported in (Gifu, 2010, 2011, 2012, 2013), was concentrated on this type
of human validation.
This paper is structured as follows. Section 2 shortly describes the
functionality of the software and the associated resources for the
Romanian language. Then, section 3 discusses two examples picked up

3
from the 2014 presidential campaign in Romania, in terms of print press,
and section 4 presents some conclusions and future work.
2 Previous work
The analysis of this type is not new. LIWC-2007, for instance, does a
similar type of investigation over texts. But this software we developed,
PDA
1
(Public Discourse Analyzer), while copying some features from
LIWC-2007, includes many new functionalities. It offers the possibility to
analyze efficiently large bodies of text and to characterize them
quantitatively and qualitatively, the results having to be as close as
possible to the analysis made by a human expert. The system offers a
global perspective over the political discourse, as well as a punctual one.
One aspect of the software presented here touches a lexical-semantic
functionality, which has some similarities with the approach used in
LIWC (Pennebaker, J. W. et al., 2001). There are, however, important
differences between the two platforms. LIWC-2007
2
is basically counting
words and incrementing counters associated with their declared semantic
classes. In the lexicon, words can be given by their long form, as a
complete string of characters, or abbreviated, in which case the sign ‘*’,
placed in a terminal position, plays the role of the universal jolly-joker,
replacing any character. For each text in the input, LIWC produces a set
of tables, each displaying the occurrences of the word-like instances of
the semantic classes defined in the lexicon, as sub-unitary values. For
each semantic class, such a value is computed as the number of
occurrences of the words corresponding to that class divided by the total
number of words in the text. It remains in the hands of the user to
interpret these figures. And there is no support for considering lexical
expressions.
The PDA software performs lemmatisation of words. This is why the
lexicon can now be declared as a collection of lemmas having the POS
categories: verb, noun, adjective and adverb. As seen, we leave out the
pronouns, numerals, prepositions and conjunctions, considered to be
semantically empty.
An entry of the lexicon has the form:
<lemma> <POS> <sem-list> (1)
1
http://nlptools.info.uaic.ro/Resources.jsp
2
www.liwc.net

4
where <sem-list> is a list of semantic classes.
Fig. 1. The PDA interface
Fig. 1. The session work in PDA

5
This means that the same lemma can appear with more than one POS and,
if needed, with different semantic interpretations. The user has the
possibility to either define an entry as a <lemma><POS><sem-list>
triple.
The second range of differences between the two platforms stays in
the user interface. In PDA, the user has an easy interactive interface,
offering a lot of services: opening a file in different formats (.txt, .doc,
.docx, .pdf), modifying/editing and saving the text, functions of
undo/redo, functions to edit the lexicon, etc. Then, the menus offer a
whole range of output visualisation functions, from tabular form to
graphical representations and to writing services. Figure 1 shows a
snapshot of the interface during a working session.
Finally, another important development was the inclusion of a
collection of formulas which can be used to make comparative studies
between different subjects. In section 3 we will present an example.
The PDA lexicon now contains over 10.000 lemmas and 33 semantic
classes presented in Table 1. For instance, social class includes a
large group of words that denote social processes, including all non-first-
person-singular personal pronouns as well as verbs that suggest human
interaction (talking, sharing), like: presidency, event, celebration. We
work now to populate the lexicon further by importing the words with
emotional load from DEX-online
3
. The semantic classes in PDA are
partially placed in a hierarchy.
Table 1. The 33 semantic classes currently included in the PDA package
code
Classes
Description and examples
1
swear
Words that harm the reputation of someone: damn,
stupid etc.
2
social
Words about social appurtenance: presidency,
election, etc. It contains next 3 classes.
3
family
clan, relative, etc.
4
friends
comrade, colleagues, etc.
5
people
adulte, citizens, etc.
6
emotional
Words with emotional effect: abused, enthusiasm, etc.
It contains next 2 classes.
7
positive
Words with positive effect: nice, popular, etc.
8
negative
Words with negative effect: aggressive, cynic, etc. It
contains next 3 classes.
9
anxiety
umpy, suspicious, etc.
10
anger
offensive, irascible, etc.
3
DEX-online is the greatest public online dictionary for Romania.

6
11
sadness
alarming, sad, etc.
12
rational
Explicative words: inventive, admissible, etc. It
contains next 5 classes.
13
intuition
explicative, decisive, etc.
14
determine
deductive, inventive, etc.
15
uncertain
confuse, ambiguous, etc.
16
certain
precise, absolutely, etc.
17
inhibition
stammer, ezitant, etc.
18
perceptive
Words related to perceptions:
hear, observes, flexible, etc. It contains next 3 classes.
19
see
tare, incandescent, etc.
20
hear
acoustic, bluntly, etc.
21
feel
sharp, tasted, etc.
22
sexual
Words with a sexual tonality:
bitch, hot, etc.
23
work
Work specific words:
adviser, organization, etc.
24
achievements
Words that reveal human accomplishments:
award, skills, etc.
25
failures
Words that reveal human errors:
mistake, unfulfillment, etc.
26
leisure
Words from the recreational domain:
dance, camping, etc.
27
home
Words related to house:
room, attic, etc.
28
financial
Words related to economy and finance:
wealth, account, etc.
29
religion
Words with a religious tonality:
Bible, God, etc.
30
nationalism
Words with a patriotism tonality:
nation, Independence, etc.
31
moderation
Words with a moderate tonality:
well, good, etc.
32
firmness
Words expressed firmly:
confident, firm, etc.
33
spectacular
Words expressed sensational:
sensational, brilliant, etc.
A special section of the lexicon includes expressions. An expression
is defined as a sequence:
<root-list> => <sem-list> (2)
where <root-list> is a list of roots of words, therefore each
optionally followed by the ‘*’ sign. Because, in principle, a root can also
be a numerical value and the semantic classes in <sem-list> are
indicated by numbers, to separate the roots section from the semantic

7
categories section we had to place a special sign (=>). Each time a
sequence of words matching the <root-list> is recognised in the text,
the counters associated with the semantic classes in the <sem-list>
are increased.
There is a hierarchical classification of partial classes expressed, as
seen, in XML:
<classes>
<class name="social" id="2"/>
<class name="family" id="3" parent="2"/>
<class name="friends" id="4" parent="2"/>
<class name="people" id="5" parent="2"/>
</classes>
The social class has three subclasses: family, friends
and people. The same situation for emotional, rational,
positive and negative classes.
For instance:
The lemma mother, the following classes are assigned: 2 = social, 3
= family.
Whenever the word or a variant of the word is detected in the input
file, all three counters, corresponding to the classes mentioned, are
incremented. In order words, the lexicon assigned to the first class is a
sum of all words/roots from subclasses.
The 33 semantic classes included now in PDA have been selected to fit
optimally with the necessities to interpret the political speech of the
Romanian presidential campaign, in 2014. Also, the user can define at
his/her will these classes and the associated lexicon.
3 The Romanian presidential elections in 2014
The articles from newspapers are a mirror of contemporary society,
rapidly to reevaluate the socio-political life. Any political text may create
inconvenience or may become useful for a power group of the moment.
For the elaboration of preliminary conclusions over the presidential
elections process, conducted in the period 3-14 November 2014 (the 2nd

8
tour of the election’s campaign) in Romania, we collected, stored and
processed electronically political texts
4
.
In essence, the program receives the input from a file, and counts
occurrences of words belonging to its defined classes. The user can notice
directly the mentioned semantic classes (and the corresponding
frequencies), as the words belonging to a selected class appear underlined
with a colour in the left screen. The user can choose a type of graphical
representation ("function", "pie" or "columns"), which give intuitive
visual perceptions on which the interpretation of discourse data can be
performed more conveniently.
PDA provides a library of comparative functions, with 2 to n different
input streams of data,
Nn
, where N is a set of natural numbers. One
stream can be either a newspaper, or only one discursive approach on a
certain topic delivered at a certain moment in time by the traced author.
To exemplify, we present below two cases in two different graph
representations:
1. In fig. 2 we see one type of graph that points out the predominant
discursive characteristics of one of the three monitored dailies. In this
example we analyse comparative differences between “x” daily and the
average of the other two dailies (“y”, “z”). We named this type of analysis
“one-to-two”, as given by Formula (3), included in the PDA
Mathematical Functions Library:
2)()(
)(
21 ,, zaverageyaverage
xaverageDiff zyx



(3)
where x, y and z are three streams; average(x), average(y) and average(z)
are the average frequencies of x, y and z over the whole stream, and the
difference is computed for each selected class. Since a difference can lead
to both positive and negative values, these particular graphs should read
as follows: values above the horizontal axis are those prevailing at the
daily x versus the daily y and the daily z, and those below the horizontal
axis show the reverse prominence. A zero value indicates equality. Our
experience shows that an absolute difference value below the threshold of
0.5% should be considered as irrelevant and, therefore, ignored in the
interpretation.
4
Editorials, published by three national publications and political speeches (in both
oral and written form), belonging to two candidates for the function of President.

9
Fig. 2. The average differences in the frequencies for each class after processing the
editorials (the second tour), between the newspapers: Evenimentul zilei versus Gândul &
Ziua.
So, the graphical representation in Fig. 2, in which Evenimentul zilei
(above Ox axis) is compared with Gândul & Ziua (below Ox axis), should
be interpreted as follows: Evenimentul zilei’s discursive interventions
have an interest more towards the working aspects (work class) than
Gândul and Ziua together, which had a argue point of view (rational
class).
2. In fig. 3 we see one type of graph that points out the predominant
discursive characteristics of one of the two presidential candidates,
remained in the second tour of elections. In this example we analyse
comparative differences between “x” and “y” candidates. We named this
type of analysis “one-to-one”, as given by Formula (4), included, also, in
the PDA Mathematical Functions Library: also:

Diffx,y
11average(x)average(y)
(4)
where x and y are two streams; average(x) and average(y) are the average
frequencies of x and y over the whole stream, and the difference is
computed for each selected class. Since a difference can lead to both
positive and negative values, this particular graph should read as follows:
values above the horizontal axis are those prevailing at the candidate x

10
versus the candidate y, and those below the horizontal axis show the
reverse prominence. A zero value indicates equality.
To exemplify, we present below a graph with two streams of data,
representing the political discourses between the two remaining
presidential candidates in the election race, the second tour of voting.
Fig. 3. The average differences in the frequencies for each class after processing the
texts from debates (second tour), between first candidate versus the second candidate.
So, the graphical representation in Fig. 3, in which candidate 1 (C1)
(above Ox axis), is compared against candidate 2 (C2) (below Ox axis),
should be interpreted as follows: C1’s discursive intervention are much
more rational than C2’s, therefore touching directly onto human common
sense, being also rather determine (cause class), while his opponent,
C2, has touched more onto social issues (social class) and he had
an emotional attitude (emotional class).
Many of the conclusions found by the program have been confirmed
by political commenter’s. Moreover, the program helped also to outline
distinctive features which brought a new, and sometimes even
unexpected, vision upon the discursive characteristics of the presidential
candidates, of the columnists and, last but not least, of the Romanian
electors, at the end of 2014.
4 Conclusions and discussions
Currently, the NLP domain has reached a scientific and technological
maturity that makes it useful in the activities carried out by researchers in
socio-humanity fields.

11
This method has a range of features that make it attractive as a tool to
assist political campaigns. It can also be rapidly adapted to new domains
and to new languages, while its interface is user-friendly and offers a
good range of useful functionalities. The program helps to outline
distinctive features which bring a new, and sometimes even unexpected,
vision upon the discursive characteristics of the political author, or of the
columnists.
In the future our intention is to include a word sense disambiguation
module in order to determine the correct senses, in context, of those
words which are ambiguous between different semantic classes belonging
to the lexicon, or between classes in the lexicon and outside the lexicon
(in which case they would not have to be counted). Also, we plan to align
the semantic classes in PDA with WordNet
5
for languages which support
this type of linguistic resource.
Acknowledgments: This work was published with the support of the
ERASMUS MUNDUS Project EMERGE (Erasmus Mundus European
Mobility with Neighbouring ReGion in the East), Action 2 – Strand 1
(2009-2013), Grant Agreement no. 2011-2576/001-001-EMA2, (Lot 8:
Moldova, Ukraine, Belarus), funded by the European Union.
References
1. Cicero. 1973. Opere alese, vol. II, Ed. Univers, Bucureşti, p. 51.
2. Cristea, D., Tufiş, D. 2002. The Romanian Linguistic Resources and
information technologies applied to Romanian language. In Ichim,
Ofelia, Olariu, Florin Teodor, Identitatea limbii şi literaturii române în
perspectiva globalizării. Academia Română, Institutul de Filologie
Română „A. Philippide”, Iaşi: Trinitas, pp. 211-234.
4. Fox, B. A. 1987. Discourse structure and anaphora. Written and
conversational English. Cambridge: Cambridge University Press.
5. Gîfu, D. 2010. The Discourse of the Print Press and the Violence of
Symbols. Analysis of an election campaign. PhD thesis. Faculty of
Philosophy and Political Studies, “Alexandru Ioan Cuza” University of
Iaşi.
6. Gîfu, D. and Cristea, D. 2011. Computational Techniques in Political
Language Processing: AnaDiP-2011 in Communications in Computer
5
http://wordnet.princeton.edu/

12
and Information Science, vol. 185 CCIS, Issue PART 2, pp. 188-195, J.J.
Park, L.T. Yang, and C. Lee (Eds.) Springer-Verlag Berlin Heidelberg.
7. Gîfu, D. 2012. Political Text Categorization in Humanities and Social
Sciences Review, Vol. 1, No. 3, Rhode Island, USA.
8. Gîfu, D. 2013. Italian political discourse analysis. Opinion and feeling
in International Journal of Communication Research, vol. 3(1),
January/March 2013, pp. 40-52.
6. Pennebaker, J. W., Francis, M. E., Booth, R. J. 2001. Linquistic
Inquiry and Word Count – LIWC200. Mahwah, NJ: Erlbaum Publishers
(www.erlbaum.com).
7. Perelman, C. et Olbrechts-Tyteca, L. 1972. Traité de l'argumentation,
Éd. de l'Institut de Sociologie de l'Université Libre de Bruxelles, p. 72.
8. Quintilian, M. F. 1971. Institutio Oratoria, II, XV, 21, Leipzig
B.S.B., Bibliotheca Scriptorum Graecorum et Romanorum Teubneriana.
9. Stevenson, N. 1995. Understanding Media Cultures: Social Theory
and Mass Communication, Londra: Sage.
11. van Dijk, T A. 1972. Textual Structures of News in the Press.
Working notes. University of Amsterdam, Department of General Literary
Studies, Section of Discourse Studies
12. Wolton, D. 1995. Les contradictions de la communication politique.
Paris: CNRS Editions.