Content uploaded by Sunny Rai
Author content
All content in this area was uploaded by Sunny Rai on Mar 29, 2022
Content may be subject to copyright.
Proceedings of the 18th International Conference on Natural Language Processing, pages 431–438
Silchar, India. December 16 - 19, 2021. ©2021 NLP Association of India (NLPAI)
#covid is war and #vaccine is weapon?
COVID-19 metaphors in India
Mohammed Abdul Khaliq, Rohan Joseph, Sunny Rai
School of Engineering Sciences
Mahindra University, Hyderabad, India.
{khaliq170568},{rohan18545}@mechyd.ac.in
sunny.rai@mahindrauniversity.edu.in
Abstract
Metaphors are creative cognitive constructs
that are employed in everyday conversation
to describe abstract concepts and feelings.
Prevalent conceptual metaphors such as WAR,
MONSTER, and DARKNESS in COVID-19
online discourse sparked a multi-faceted de-
bate over their efficacy in communication, re-
sultant psychological impact on listeners, and
their appropriateness in social discourse. In
this work, we investigate metaphors used in
discussions around COVID-19 on Indian Twit-
ter. We observe subtle transitions in metaphor-
ical mappings as the pandemic progressed.
Our experiments, however, didn’t indicate any
affective impact of WAR metaphors on the
COVID-19 discourse.
1 Introduction
Metaphors are cognitive artefacts to anchor one’s
thoughts and navigate a situation whether it is
social, political or even financial. Conceptual
metaphors restructure an abstract domain in terms
of a relatively concrete domain, influencing how
we perceive reality [Lakoff and Johnson,1980].
Consider the conceptual metaphor, “DREAMS
are BUTTERFLIES”. Here, the abstract domain,
DREAMS is mapped to a more concrete domain
as that of BUTTERFLIES evoking meanings such
as vibrant and delicate. Linguistic metaphors are
the manifestations of these conceptual mappings
in text. For instance, Her eyes were full of vibrant
dreams.
India reported the first case of COVID-19 infec-
tion in January, 2020 [Andrews et al.,2020]. Soon
after that, various control measures including the
restriction on international travel, screening of air
passengers and institutional quarantine were im-
plemented to curb the infection. The government
of India imposed the first nationwide lockdown
1
1
‘Coronavirus in India: 21-day lockdown begins; key high-
lights of PM Modi’s speech’, Business Today (Mar 25, 2020).
Available at LINK
from Mar 25, 2020 to Apr 14, 2020 as a preventive
measure to curb COVID-19.
Various conceptual metaphors with source do-
mains such as WAR (defeat the virus), HAM-
MER/LANDSCAPE (flatten the curve) and even
MONSTER (grappling with virus) were used to
communicate state’s guidelines as well as reactions
to situations arising due to the pandemic [Ru
˜
ao
and Silva,2021]. In the mapping, COVID-19 is
WAR, healthcare staff have been reconceptualised
as warriors, and the citizens as soldiers fighting
unitedly against the enemy COVID-19. Flusberg
et al. [2018] advocate the use of domain WAR to
deliver urgent communication infused with motiva-
tion. Consider another phrase “Covid 2.0: Threat
of an administrative cytokine storm building up
in India”
2
. Here, administration’s RESPONSE is
being compared to a biological phenomenon, CY-
TOKINE STORM to emphasize the lack of atten-
tion to ground-reality while drafting COVID-19
protocols. This mapping does bundle subtle as-
pects such as failure to identify key points of action,
overly restrictive protocols and its resultant unde-
sired effect on citizens.
Another direction of research on COVID-19
metaphors reflects on the appropriateness of these
cognitive constructs in COVID-19 discourse. There
is widespread discontentment against re-imagining
the pandemic using source domains such as WAR
or MONSTER which may negatively manipulate
the understanding of the society. World Emergency
COVID-19 Ethics (WeCope) Committee advise
against the use of WAR metaphors as it instills
fear amongst masses and leads to stigmatization
towards those who do not respect the guidelines
[WeCope,2019]. H
¨
oijer [2011] calls out the use of
metaphors as an instrument by the state to defend
its actions and policies during the pandemic. Por-
2
Blog by Samir Shukla, Times of India (Mar 30, 2021).
Available at
https://timesofindia.indiatimes.
com/blogs/science-nomad/
431
traying the pandemic as a WAR legitimizes state-
imposed violence and excessive control. Kahamb-
ing [2021] warns against the portrayal of COVID-
19 virus as Mother Nature’s way to clean the planet.
Sabucedo et al. [2020] further demonstrated the ill-
effect of commonly used violent source domains
such as WAR, MONSTER in COVID-19 discourse
on public health. Semino [2021] thus emphasizes
the need to reframe COVID-19 metaphors. Rohela
et al. [2020] speak in favour of correcting the WAR
narrative in COVID-19 discourse using substitute
domains such as CRICKET and DANCE. Inspired
from this debate, we study the metaphors embed-
ded in Indian tweets posted during the pandemic.
The main contributions of this study are:
•
We manually identify COVID-19 conceptual
metaphors used in headlines of major In-
dian newspapers published from March’20
to May’21.
•
We detect linguistic metaphors embedded in
Indian tweets on COVID-19 posted between
Mar’2020 to Jul’2021 by fine tuning BERT
model.
•
Using diachronic embeddings, we detect the
transition in manifestations of conceptual
metaphors as the pandemic progressed.
•
We study the hypothesis if the conceptual
metaphor WAR has an affective influence on
COVID-19 discussion on Twitter.
The rest of the paper is organized as follows. We
discuss the prior works on conceptual metaphors in
context of the ongoing global pandemic in Section
2. We describe data collection and the procedure
to frame different conceptual metaphors in Section
3. Section 4discusses the evolving interpretation
of mappings as well as the role of WAR metaphors
on COVID-19 discourse. We conclude our work in
Section 5.
2 Related Work
Wicke and Bolognesi [2020] studied the source
domain WAR and how its pervasive nature when
describing diseases, plays a role in the COVID-
19 dialogue. Prior research discusses the role
of conceptual metaphors in moulding public per-
ception in India [Rohela et al.,2020,Wagener,
2020]. Das [2020] describes the crisis and wrath
faced by marginalized sections of society due to
WAR centered analogies by the Government of In-
dia. The readers are encouraged to refer [Rai and
Chakraverty,2020] to know more about metaphors
and theories.
Our work differs from the existing works in mul-
tiple ways. First, our work focuses on the COVID-
19 metaphors of India. We detect metaphorical
tweets and also label the underlying conceptual
mapping. We study the transition in the linguistic
metaphors as the pandemic progressed. Taking in-
spiration from WEAT [Caliskan et al.,2017], we
measure affective influence of metaphorical tweets
on COVID-19 discourse as well as see if WAR
metaphors indeed present a grimier picture. To the
best of our knowledge, this is the first computa-
tional approach designed to understand metaphori-
cal themes in India during the pandemic.
3 Discovering Metaphors of COVID-19
3.1 Dataset
Twitter is a micro blogging platform, used widely
during the pandemic to express one’s feelings and
seek help. We extracted Indian tweets on COVID-
19 posted between March’20 to July’21 using
snscrape
3
library. A tweet is considered a COVID-
19 tweet if it has at least one COVID-19 related
hashtag such as #coronavirus, #covid19, #quaran-
tine, #covid 19, #vaccine, #TogetherAgainstCovid
etc. To ensure that the extracted tweets are from
India, we check if the location attribute within the
tweet object pulled by snscrape contains ”India”.
Our dataset comprises of over 1.3Mtweets.
3.2 Filtering literal tweets
To filter out literal tweets, we fine tuned a BERT
model [Devlin et al.,2018]. Two human annota-
tors were asked to tag a random subset of collected
tweets into categories metaphor and not metaphor,
for the task of finetuning. Both annotators are un-
dergraduate students aged between
19 −22
, pro-
ficient in English with sufficient knowledge of In-
dian society. The guidelines shared with annotators
to identify metaphors in tweets is as described by
Pragglejaz group [Group,2007]. Below are the
steps:
•
Read the text to get a general understanding
of the meaning
• Determine the lexical units
3https://pypi.org/project/snscrape/
432
–
Establish the contextual meaning of the
unit
–
Determine if it has a more basic meaning
•
Does the contextual meaning contrast with
the basic meaning but can it be understood in
comparison with it?
• If yes, mark the unit as metaphorical.
A total of
3.7K
tweets were tagged by annota-
tors, out of which
1.8K
were marked as metaphor-
ical. We obtained Cohen’s kappa of
0.719
on a
common sample of
100
tweets indicating good re-
liability of annotation. The hand annotated dataset
of tweets is available at link4.
We split this dataset into a training set with
3006
tweets, test and validation sets with
376
tweets
each. On finetuning BERT for
25
epochs with a
learning rate of
2e−5
, we obtained an accuracy of
74.4%
on the validation set. On the test set, we
achieved accuracy of
72.6%
with a precision of
77% and recall of 67%.
We used this finetuned model to identify
metaphorical tweets from the dataset collected in
Sec. 3.1. Out of almost
1.3M
tweets, the sys-
tem predicted
264K
tweets as metaphorical. We
hereafter indicate this set of metaphorical tweets as
M0.
3.3 Framing the source domain
The next task is to derive a list of conceptual
metaphors used in COVID-19 metaphorical tweets
from India. #ReframeCovid
5
is one such ongo-
ing open-source work which collects metaphorical
mappings present in global COVID-19 tweets and
related media.
For our study, we created a list of metaphori-
cal mappings
S0
(of the form SOURCE DOMAIN
is TARGET DOMAIN) inspired from #Reframe-
Covid along with the manual analysis of major In-
dian newspaper headlines on COVID-19 published
during Mar’20-May’21. Few examples are VAC-
CINE is SHIELD, COVID-19 is TEACHER and
PANDEMIC is SPEEDBREAKER. The complete
list of mappings is available at link4.
Prior works [Choi and Lee,2019,Wicke and
Bolognesi,2020] used websites
6
to extract lexi-
cal units that they then use to frame the source
4https://github.com/makflakes/
Covid-Metaphors- of-India/tree/main/data
5https://sites.google.com/view/
reframecovid/initiative
6www.relatedwords.org
domains. However, we found that the these lex-
ical units are not used commonly in Indian En-
glish. We thus use pretrained word2vec embed-
dings [Mikolov et al.,2013] to expand the set of
relatable lexical units/concepts close to a source
domain
s∈S0
. We train a word2vec skip-gram
[Mikolov et al.,2013] model on the 264K tweets
in
M0
to derive the lexical units . We define lower
and upper thresholds for cosine similarity to filter
out overly specific as well as generic concepts. The
thresholds are decided empirically. The set
Ps
de-
notes the set of lexical units that we use to frame
source domain
s∈S0
. Below are few lexical units
from the set Pwar:
fight, battle, defeat, enemy, soldier, menace, bi-
ological war, weapon, battlefield, soldier, defend,
warfare, bio war, unseen enemy, hero, confronta-
tion, army, combat, biowar, invasion, biowarfare,
destruction, war, attack, superpower, destruction,
fighting, standoff, invisible enemy, invade, invasion
We manually filter the list to make it contextually
suitable for COVID-19 discourse. For instance, the
above list for the source domain WAR comprises
of words such as menace,hero,superpower which
was used in literal sense while discussing the pan-
demic. Therefore, these words are removed from
the list. This forms the basis for identifying un-
derlying source domains in Indian metaphorical
tweets.
3.4 Identifying Conceptual Metaphors in
Tweets
In this section, we describe our approach to identify
the inherent conceptual mapping that is, TARGET
DOMAIN is SOURCE DOMAIN in the tweets.
3.4.1 Labeling Source domain
We categorize a tweet
t
to source domain
s∈S0
if
t
consists a word
w∈Ps
. There is a possibility
that a tweet may have words related to two or more
source domains. For our analysis, we have elimi-
nated these tweets and will only be utilising tweets
that uniquely indicate a particular source domain.
Below are few example tweets:
“ A
storm
is coming. Brace yourselves. Impact
on Indian economy will be severe. I have started
studying and will write a detailed article on it. Will
publish soon. #coronavirusindia”. - COVID-19 is
STORM
“In a #World divided by #religion, greed and
433
Table 1: Top-10 Source Domains
Source Domain #Tweets
WAR 48415
MONSTER 2884
SUCCESS/CHALLENGE 1382
LESSON/TEACHER 1252
STORM 1213
DARKNESS 1080
PUNISHMENT/BANE 851
PRISON 851
LUXURY 602
CATALYST 542
SAVIOR 486
SHIELD/BARRIER 426
inflated egos, it took an
invisible
virus to instill a
common fear. And we still believe that #human be-
ings are the most intelligent species on this planet.
#coronavirus” - COVID-19 is DARKNESS
“Janata Curfew on 22nd March 2020 from 7
AM to 9 PM AND ”Ghantanad” on 5 PM, which
will help Indians to fight against the corona viruse.
It will help to kill the
devil
”- COVID-19 is MON-
STER
We list the Top-10 source domains on the ba-
sis of their volume in
M0
in Table. 1. It may be
noted, this list is derived using
s∈S0
. It is thus
possible that there are undetected source domains
s /∈S0
with metaphorical tweets in our dataset.
From Table. 1, we observe that WAR is the most
often used source domain to describe COVID-19 re-
lated events. Source domains such as MONSTER,
CHALLENGE, LESSON, STORM also contribute
significantly to the discourse.
3.4.2 Assumption regarding target domain
Since the tweet extraction process focused only
on tweets with COVID-19 related hashtags, it is
safe to assume that all tweets are inherently de-
scriptors of COVID-19 and related dialogue. We
initially considered segregating tweets with vac-
cine related hashtags to the target domain VAC-
CINE. We discovered a total of 3701 metaphorical
tweets on VACCINE. On careful analysis, we dis-
covered that tweets tagged with VACCINE related
hashtags, were also essentially reconceptualizing
COVID-19/PANDEMIC. Few such tweets are pro-
vided below.
“Another
deadly
wave of Covid19 is ravaging
countries including India Stricter observance of
anti Covid protocol and stepping up vaccination
manifold are urgently called for to face the cri-
sis #tkan #vaccine” - VIRUS/COVID-19 is MON-
STER
“Got vaccinated today with first dose of in-
digenously developed #Covaxin Thank you naren-
dramodi. Thank you all the scientists who worked
hard to invent the vaccine in record time. To-
gether India will
defeat
COVID-19.” - COVID-19
is WAR/VIRUS is ENEMY
Due to the plentiful presence of such tweets in
the VACCINE targeted set, we decided to go ahead
with COVID-19 as the sole target domain for fur-
ther analysis.
4 COVID-19 Metaphors of India
4.1 Evolving Conceptual Mappings
As the pandemic progressed, the conceptual map-
pings describing COVID-19 also evolved. Consider
the topic of VACCINE, which was initially concep-
tualized as a WEAPON
7
to decimate the enemy,
COVID-19 virus. Later, VACCINE evolved into a
PASSPORT/TICKET
8
to freedom which allowed
unrestricted movement and gradually, it metamor-
phosed to LUXURY
9
which was rare and accessi-
ble to only few.
In this section, we study the evolving concep-
tualization of COVID-19 through the notion of
semantic shift. To compute semantic shift, the
standard approach is to first slice a corpus with
respect to time. The granularity for time slicing
may vary depending on the problem statement.
For our analysis, we consider the duration (a)
t0
:
during lockdown i.e. Mar’20 to Jun’20 (b)
t1
:
post lockdown i.e. July’20 to Oct’20 and (c)
t2
:
second wave i.e. Mar’21 to Jun’21. Given cor-
pora
C= [Ct0, Ct1, Ct2]
, the task is to analyse
the change if any in semantic neighbourhood of
COVID-19. Here,
Cti
indicates the set of metaphor-
ical tweets posted during the time interval ti.
On slicing the set of metaphorical tweets
M0
from Sec. 3.2,
Ct0
has 136K tweets,
Ct1
has 39K
tweets and
Ct2
contains 65K tweets. We learn word
embeddings using word2vec skip-gram architec-
ture [Mikolov et al.,2013] for the first phase using
7
The Hindu, May 26, 2021. ”Vaccination is our only
weapon” available at LINK
8
The Diplomat, July 12, 2021. ”Vaccine Passports: Ticket
to Freedom or Path to a Divided World? ” available at LINK
9
Quartz India, Aug 2, 2021. ”India’s vaccine supply is a
curious mix of abundance and shortage” available at LINK
434
(a) Lockdown (b) Post Lockdown
(c) Second Wave
Figure 1: t-SNE representation of diachronic word embeddings
time-specific corpora
Ct0
. Here, the vectors are ran-
domly initialized. For next two phases, we update
the embeddings initialized with the embeddings
learnt from the previous phase. In order to com-
pare word vectors from different time-periods, we
align word vectors to the same coordinate axes us-
ing orthogonal Procrustes [Hamilton et al.,2016].
4.1.1 Analysis
To analyze the semantic shift in the conceptualiza-
tion of COVID-19, we plot the t-SNE visualization
[van der Maaten and Hinton,2008] in Fig. 1for all
phases.
Fig. 1-a visualises the semantic concepts re-
lated to COVID-19 based on tweets posted in
t0
phase. Fear and determination are reflected from
the metaphors such as scare, fear, panic, worry,
deadly, dangerous and combat, win, threat defeat,
protect respectively used in this phase. COVID-19
is conceptualized as MONSTER, and even as an
OBSTACLE/GAME (challenge, overcome, strike,
tackle). This overview is consistent with the mixed
feelings of fear and hope in the early stages of
COVID-19 in India.
The post lockdown embeddings are aligned on
t0
embeddings. We present covid related concepts
for this period in Fig. 1-b. Metaphors such as fight,
call, win, race are getting closer to COVID-19
indicating the positive attitude. MONSTER related
words such as fear, scare, deadly are moving away.
BARRIER metaphors such as shield and other units
such as nature,shame,ruin start to appear in this
phase.
Concepts related to covid from the Second wave
embeddings aligned on
t1
are depicted in Fig. 1-c.
Defensive WAR metaphors such as failure,lose,
tough and shield are getting closer. DARKNESS
related metaphors such as grim,vanish,invisible
can be seen. MONSTER related metaphors have
come closer when compared to Fig. 1-b. Orienta-
tional metaphors such as fall, and surge are also
present.
4.2 Linguistic Metaphors
To identify linguistic metaphors of each phase, we
extracted the top-1000 words closest to covid19
from each of these phases. We manually go through
this list to identify linguistic metaphorical units.
Relevant tweets were retrieved when needed to sub-
stantiate the metaphoricity and rule out the literal
use. We present these linguistic metaphors in order
of increasing distance from covid19 in Table 2.
The foremost observation is the pronounced pres-
ence of WAR metaphors across all phases. This
435
Table 2: Linguistic Metaphors of COVID-19
Phase Linguistic metaphors (w.r.t cosine distance from COVID-19 in vector space)
Lockdown (t0)
fight, indiafightscorona, battle, crisis, deadly, break, war, defeat, support, control, win, combat,
attack, force, team, protect, hide, coronawarrior, threat, kill, dangerous, suffer, prepare, impact,
isolation, hit, border, hell, solution, strategy, survive, fighting, coronafighter, unite, develop,
strength, scary, destroy, isolate, disaster, tackle, scare, beat, deadly virus weapon, soldier, fighter,
shut, shame, lesson, cover, enemy, scared, win battle, win war, danger, devastating, catch, surge,
tracking, deep, victory, vulnerable, evil, cut, surpass, giant, expose, break chain, boom, unlock,
push, hail, chain, flatten curve, heal
Post-Lockdown (
t1
)
indiafightscorona, break, stand, block case, battle, line, strong, support, war, combat, save,
warrior, frontline, duty, fighter, leader, play, win, strategy, fear, hit, control, covidwarrior, scary,
deadly, attack, base, push, tackle, brave, hell, team, force, defeat, game, fall, race, power, action,
united, build, struggle, beat, dangerous, kill, strike, panic, powerful, peak, danger, stage, crusader,
frontline warrior, tough time, scared, deadly virus, deep, crisis, hide, throw, win battle, lose life,
rage, threat, grim, nightmare, block, havoc, unlock, fire, fighting, flood, wall, victory, impact,
kick, boost, storm, invisible, weapon, disaster, shoot
Second Wave (t2)
battle, support, deadly, crisis, strong, win, suffer, defeat, hit, indiafightscovid, handle, dangerous,
safety, warrior, control, strength, protect, overcome, fear, difficult, attack, tough time, hard,
scary, war, kill, struggle, win battle, tough, scare, pain, beat, peak, panic, strike, grim, catastro-
phe, save life, unite fightcorona, hell, shame, difficult time, lose life, combat, frontline worker,
stay united, devastating, disaster, wake, hero, breakthechain, powerful, shift, destroy, fighting,
deadly virus, indiafightsback, blast, tackle, seek, shocking, dip, weapon, force, rule, front-
line warrior, game, chance, strategy, decline, hang, target, lightly, enemy, border, wish speedy,
lose battle, threat, loss, shall pass, push, catch, build, breach, blame, player, rage, bio bubble,
shock, frontline, hit hard, nightmare, gloom, danger, tsunami, tear, kick, casualty, terrible, brutal,
lethal, mark, pressure, devastate, devastation,
includes enemy, fight, defeat, attack, weapon, bat-
tle, casualty etc. We further note the presence
of domains including MONSTER (deadly, scary,
giant, fear), GAME (team, race, tackle, push,
player, strike), STORM/DISASTER (crisis, dev-
astate, havoc, flood, hail, tsunami), DARKNESS
(hide, cover, nightmare, gloom, grim, invisible),
HAMMER (beat, flatten curve, hit, impact) and
ACCIDENT (chance, rage, shock). There are also
other metaphors such as dip, surge, blast, tear, kick,
build, shame, hell, evil, heal, chain, deep in the
corpus.
During
t0
phase, the closest words are fight, bat-
tle, crisis, deadly, break, war, protect, hide etc. It
may be noted that India had only few reported cases
of COVID-19 in comparison to
t1
phase. Never-
theless, the metaphors are relatively grim and fear
inducing. The fear of unknown and the lack of con-
fidence on Indian healthcare might be the reason
behind these overly gloomy tweets.
We note increased use of GAME metaphors in
t1
phase when compared to other two phases. It
may be noted that Post-Lockdown is the phase
where India faced the first wave of COVID-19.
COVID-19 is discussed using concepts such as
race,action,kick,team and block. Moreover, even
WAR metaphors are used in more authoritative fash-
ion when compared with the Second Wave phase.
This indicates the transition in lexical manifesta-
tions of WAR metaphors while discussing COVID-
19. Metaphors such as duty, strategy, push, tackle,
brave, fighter, crusader convey a sense of control.
The metaphors used in this phase indicate a more
confident and controlled reaction to the pandemic
in comparison to t0and t2phases.
We see the highest volume of metaphors in
Second Wave phase. There are more negative
metaphors such as battle, deadly, crisis, suffer, de-
feat, dangerous closer to COVID-19 when com-
pared with the previous two phases. There are also
increased occurrence of DISASTER/DARKNESS
metaphors such as grim, catastrophe, devastating,
disaster, panic, nightmare, gloom, danger, tsunami
etc. We also note metaphors such as breach, blame,
rage, pressure from GAME domain indicating the
shift in meaning of GAME metaphors. There is a
clear difference in the underlying emotional tone
of metaphors when compared with
t1
phase. The
first wave (
t1
) definitely saw a controlled strategy
with manageable COVID-19 cases whereas the
second wave (
t2
) witnessed more suffering, panic
and lack of control which is also evident from the
metaphors.
4.3 Impact of WAR metaphors on COVID-19
online discourse
To better understand the role of WAR metaphors
in COVID-19 discourse, we analyse if metaphors
based on WAR mapping indeed paint an overly
436
grim picture contrary to the true reality of COVID-
19 in India. We take inspiration from Word Em-
bedding Association Test (WEAT) proposed by
Caliskan et al. [2017] to identify the polarity of as-
sociativity between WAR metaphors and COVID-
19 concepts. We define our hypothesis
Ha
as “
WAR has an affective influence if COVID-19 con-
cepts show associativity towards negative or posi-
tive attribute sets.” The null hypothesis
Ho
there-
fore is “ WAR metaphors have no affective influ-
ence on the meaning of COVID-19 concepts.”
Let
Do
be the dataset of non-metaphorical In-
dian tweets as predicted by the fine-tuned BERT
model in Sec. 3.2. Let
Ds
be the set of predicted
metaphorical tweets belonging to source domain
s∈S0
. For our analysis, we first learn word em-
beddings using skip-gram word2vec model on the
dataset
Do
. These representations will serve as the
baseline for our analysis. We fine tune the learnt
embeddings using
Ds
to capture if there is a change
in the meaning of COVID-19 concepts due to the
source domain s.
For analysis, let
X
be the set of COVID-19 tar-
get words, and
P
,
N
be the attribute sets namely
positive and negative respectively. We define
δ(X,P,N)
as the differential association of the
target words embeddings for
x∈X
trained on
Do
and
Do+Ds
with the attribute sets
P
and
N
as in
eq. 1.
δ(X,P,N) = X
x∈X
f(~xo,P,N)−X
x∈X
f(~xs,P,N)
(1)
where
f(x, P,N) = µp∈Pcos(x, p)−µn∈Ncos(x, n)
•~xo
refers to the embeddings for
x∈X
learnt
on Do
•~xs
refers to the embeddings for
x∈X
fine
tuned on Ds
•µindicates mean,
•cos(~a,~
b)
denotes the cosine similarity be-
tween the vectors ~a and ~
b.
Each word in sets
X,P
and
N
has occurred
at least 20 times in both corpus and are provided
below:
X
=
{
covid, corona, virus, lockdown, pandemic,
coronavirus, health, hospital }
P
=
{
hope, faith, strength, unite, support, care,
survive, recover}
N
=
{
death, panic, struggle, concern, stress,
chaos, shortage, oxygen}
Using permutation test for sampling, we discov-
ered that the domain WAR is more close to attribute
set
P
with p-value of
0.98
. We reject our hypoth-
esis
Ha
due to high p-value. It is thus not evi-
dent from our experiments that WAR metaphors
have statistically significant affective influence on
COVID-19 domain. We further performed this
analysis specifically for tweets posted during
t2
phase. However, our experiments did not reveal
any affective influence exercised due to metaphors
on the understanding of COVID-19. In future, we
aim to design extensive experiments to understand
the affective influence of different metaphorical
themes on COVID-19 discourse.
5 Conclusion
Collecting data for any figurative text related task
is a big challenge. Through this study, we release
a hand-annotated set of
3.7K
Indian tweets for
metaphor related research. A wide variety of con-
ceptual mappings were used in Indian newspapers
while reporting COVID-19 situation in India. Nev-
ertheless, we see a handful of these conceptual
mappings in Indian tweets. WAR, MONSTER,
DARKNESS and GAME are the most prominent
conceptual metaphors in Indian tweets. Our results
reveal the shift in the use of conceptual metaphors
as the pandemic progressed. Despite intense dis-
cussions on the appropriateness of the conceptual
mappings used during the pandemic, it is not evi-
dent from our experiments if WAR indeed led to
an overly negative understanding of COVID-19 in
India.
References
MA Andrews, Binu Areekal, KR Rajesh, Jijith Krish-
nan, R Suryakala, Biju Krishnan, CP Muraly, and
PV Santhosh. First confirmed case of covid-19 in-
fection in india: A case report. The Indian Journal
of Medical Research, 151(5):490, 2020.
Aylin Caliskan, Joanna J Bryson, and Arvind
Narayanan. Semantics derived automatically from
language corpora contain human-like biases. Sci-
ence, 356(6334):183–186, 2017.
Won-Gyu Choi and Kyung-Soon Lee. Conceptual
representation for crisis-related tweet classification.
2019.
437
Shreshtha Das. India’s war on covid-19: how the gov-
ernment is turning marginalised citizens into sus-
pected enemies and criminals. South Asia@ LSE,
2020.
Jacob Devlin, Ming-Wei Chang, Kenton Lee, and
Kristina Toutanova. Bert: Pre-training of deep bidi-
rectional transformers for language understanding.
arXiv preprint arXiv:1810.04805, 2018.
Stephen J Flusberg, Teenie Matlock, and Paul H
Thibodeau. War metaphors in public discourse.
Metaphor and Symbol, 33(1):1–18, 2018.
Pragglejaz Group. Mip: A method for identifying
metaphorically used words in discourse. Metaphor
and symbol, 22(1):1–39, 2007.
William L Hamilton, Jure Leskovec, and Dan Ju-
rafsky. Diachronic word embeddings reveal sta-
tistical laws of semantic change. arXiv preprint
arXiv:1605.09096, 2016.
Birgitta H¨
oijer. Social representations theory: A new
theory for media research. Nordicom review, 32(2):
3–16, 2011.
Jan Gresil S Kahambing. Metonymies, metaphors
and/or language reconsiderations for sustainability
during covid-19. Journal of Public Health (Oxford,
England), 2021.
George Lakoff and Mark Johnson. Conceptual
metaphor in everyday language. The journal of Phi-
losophy, 77(8):453–486, 1980.
Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey
Dean. Efficient estimation of word representations
in vector space. arXiv preprint arXiv:1301.3781,
2013.
Sunny Rai and Shampa Chakraverty. A survey on com-
putational metaphor processing. ACM Computing
Surveys (CSUR), 53(2):1–37, 2020.
Pallavi Rohela, Anant Bhan, Divya Ravindranath,
Devi Leena Bose, and Soumitra Pathare. Must there
be a “war” against coronavirus. Indian J Med Ethics,
pages 222–226, 2020.
Teresa Ru˜
ao and S´
onia Cristina Melo Silva. Strate-
gic science communication: the “flatten the curve”
metaphor in covid-19 public risk messaging. 2021.
Jos´
e-Manuel Sabucedo, M´
onica Alzate, and Domenico
Hur. Covid-19 and the metaphor of war (covid-19 y
la met´
afora de la guerra). International Journal of
Social Psychology, 35(3):618–624, 2020.
Elena Semino. “not soldiers but fire-fighters”–
metaphors and covid-19. Health Communication,
36(1):50–58, 2021.
Laurens van der Maaten and Geoffrey Hinton.
Visualizing data using t-sne. Journal of Ma-
chine Learning Research, 9(86):2579–2605,
2008. URL http://jmlr.org/papers/v9/
vandermaaten08a.html.
Albin Wagener. Crushed by the wheels of industry:
War, heroes, and domestic recolonization in the time
of covid-19. Postdigital Science and Education, 2:
576–580, 2020.
World Emergency COVID19 Pandemic Ethics Com-
mittee WeCope. A call to cease the use of war
metaphors in the covid-19 pandemic. 2019. URL
https://eubios.info/yahoo_site_admin/
assets/docs/WeCope_Call_to_Cease_the_
War_Metaphor.167174242.pdf.
Philipp Wicke and Marianna M. Bolognesi. Fram-
ing covid-19: How we conceptualize and discuss
the pandemic on twitter. 2020. doi: 10.1371/
journal.pone.0240010. URL https://doi.org/
10.1371/journal.pone.0240010.
438
