Annotating For Hate Speech The MaNeCo Corpus And Some Input From Critical Discourse Analysis

Annotating for Hate Speech: The MaNeCo Corpus and Some Input from
Critical Discourse Analysis
Stavros Assimakopoulos, Rebecca Vella Muskat, Lonneke van der Plas, Albert Gatt
University of Malta
Institute of Linguistics and Language Technology, Msida MSD 2080, Malta
{stavros.assimakopoulos, rebecca.vella, lonneke.vanderplas, albert.gatt}@um.edu.mt
Abstract
This paper presents a novel scheme for the annotation of hate speech in corpora of Web 2.0 commentary. The proposed scheme is
motivated by the critical analysis of posts made in reaction to news reports on the Mediterranean migration crisis and LGBTIQ+ matters
in Malta, which was conducted under the auspices of the EU-funded C.O.N.T.A.C.T. project. Based on the realization that hate speech
is not a clear-cut category to begin with, appears to belong to a continuum of discriminatory discourse and is often realized through the
use of indirect linguistic means, it is argued that annotation schemes for its detection should refrain from directly including the label
’hate speech,’ as different annotators might have different thresholds as to what constitutes hate speech and what not. In view of this, we
suggest a multi-layer annotation scheme, which is pilot-tested against a binary ±hate speech classification and appears to yield higher
inter-annotator agreement. Motivating the postulation of our scheme, we then present the MaNeCo corpus on which it will eventually
be used; a substantial corpus of on-line newspaper comments spanning 10 years.
Keywords: hate speech, annotation, newspaper comments, corpus creation
1. Introduction
While the discussion of automatic hate speech detection
goes back at least two decades (Spertus, 1997; Greevy and
Smeaton, 2004), recent years have witnessed a renewed
interest in the area. This is largely due to the prolifera-
tion of user-generated content as part of Web 2.0, which
has given rise to continuous streams of content, produced
in large volumes over multiple geographical regions and
in multiple languages and language varieties. As a result,
the exponential increase in potential sources of hate speech
in combination with the continuous introduction of legisla-
tion and policy-making aiming specifically at regulating the
phenomenon across a number of countries (Banks, 2010)
clearly necessitates the development of reliable automatic
methods for detecting hate speech online that will comple-
ment human moderation.
From a natural language processing (NLP) perspective,
treatments of hate speech focus mainly on the problem of
identification (Fortuna and Nunes, 2018). Thus, given a
span of text, the task is to identify whether it is an in-
stance of hate speech or not. This makes the problem a
case of binary classification (±hate speech), which in turn
makes it amenable to treatment using a variety of classifica-
tion methods. These supervised learning techniques require
pre-labelled training data, consisting of manually annotated
positive and negative examples of the class(es) to be iden-
tified, to learn a model which, given a new instance, can
predict the label with some degree of probability. In this
setting, the most common features traditionally used for
hate speech classifiers are lexical and grammatical features
(Schmidt and Wiegand, 2017; Fortuna and Nunes, 2018),
with more recent approaches making use of neural network
models relying on word embeddings (Badjatiya et al., 2017;
Agrawal and Awekar, 2018).
In this paper, we concentrate on the particular issues that
one has to take into consideration when annotating Web 2.0
data for hate speech. The unique angle of our perspective
is that it is informed by data-driven research in the field
of Critical Discourse Analysis (CDA), a strand of applied
linguistics that has for long dealt with the ways in which
language is used to express ideologically-charged attitudes,
especially in relation to discrimination. For some of us the
interest in the area of hate speech detection in the Web 2.0
era was originally sparked through our involvement in the
C.O.N.T.A.C.T. project which specifically targeted online
hate speech from a CDA point of view (Assimakopoulos
et al., 2017). As a matter of fact, the ongoing compila-
tion of the Maltese Newspaper Comments (MaNeCo) cor-
pus, which this paper additionally launches, started off as
a potential extension to the much smaller corpus that was
compiled for the purposes of in-depth qualitative analysis
undertaken under the auspices of C.O.N.T.A.C.T. (Assi-
makopoulos and Vella Muskat, 2017; Assimakopoulos and
Vella Muskat, 2018). Against this backdrop, the ensuing
discussion focuses on the challenges faced while annotat-
ing this corpus, which we are treating herein as a pilot for
the development of the scheme used to annotate samples of
MaNeCo, and which, given MaNeCo’s special characteris-
tics, is expected to lead to more reliable datasets that could
be used for future training and testing models of automatic
hate speech detection. Thus, our main contributions are the
following:
• A description of the challenges encountered when an-
notating hate speech
• An annotation scheme that aims to address these chal-
lenges
• A pilot dataset of on-line newspaper comments on mi-
gration and LGBTIQ+ matters, with multi-level anno-
tations (described in Section 3.).1
1
The pilot dataset is available from the authors upon request.
arXiv:2008.06222v1
[cs.CY]
14
Aug
2020

• The MaNeCo corpus: a substantial corpus of Maltese
newspaper comments spanning 10 years.2
2. Background
When it comes to automatic hate speech detection, a suf-
ficiency of training data with high-quality labelling is a
crucial ingredient for the success of any model. Even so,
previous studies “remain fairly vague when it comes to the
annotation guidelines their annotators were given for their
work” (Schmidt and Wiegand, 2017, p.8). A review of the
relevant literature reveals that the majority of previous at-
tempts to annotate Web 2.0 data for hate speech involves
simple binary classification into hate speech and non-hate
speech (Kwok and Wang, 2013; Burnap and Williams,
2015; Djuric et al., 2015; Nobata et al., 2016). There are
of course notable exceptions where the annotation scheme
involved was more or less hierarchical in nature. For ex-
ample, in Warner and Hirschberg (2012), annotators were
tasked with classifying texts on the basis of whether they
constitute hate speech or not, but were additionally asked
to specify the target of said speech in the interest of dis-
tinguishing between seven different domains of hatred (e.g.
sexism, xenophobia, homophobia, etc). Then, acknowledg-
ing that hate speech is a subtype of the more general cate-
gory of offensive language and can thus often be conflated
with it, Davidson et al. (2017) asked annotators to label
tweets in terms of three categories: hate speech, offensive
but not hate speech, or neither offensive nor hate speech.
Along similar lines, Zampieri et al. (2019) introduce an ex-
plicitly hierarchical annotation scheme that requested an-
notators to code tweets on three consecutive levels: (a) on
whether they contain offensive language; (b) on whether the
insult/threat is targeted to some individual or group; and (c)
on whether the target is an individual, a group or another
type of entity (e.g. an organization or event). Finally, an
annotation framework which, like the one proposed here,
takes into account the intricate nature of hate speech was
developed by Sanguinetti et al. (2018). Here, annotators
were asked to not only provide a binary classification of
Italian tweets as hate speech or not, but also to grade their
intensity on a scale from 0 to 4, and indicate whether each
tweet contains ironical statements or stereotypical repre-
sentations as well as how it fares in terms of aggressiveness
and offensiveness.
Despite the apparently increasing interest in the area and
the development of all the more sophisticated annotation
methodologies, a major cause for concern when it comes
to annotations used for model training and testing is that
reliability scores are consistently found to be low pretty
much across the board (Warner and Hirschberg, 2012; No-
bata et al., 2016; Ross et al., 2016; Tulkens et al., 2016;
Waseem, 2016; Bretschneider and Peters, 2017; Schmidt
and Wiegand, 2017; de Gibert et al., 2018). This effec-
tively suggests that, even in the presence of more detailed
guidelines (Ross et al., 2016; Malmasi and Zampieri, 2018;
2
Samples of this corpus can be made available upon request.
A full release is expected in future, pending licensing agreements
with the donors of the MaNeCo data, which will need to cover sen-
sitive data such as comments written by online users, but deleted
by the moderators of the newspaper portal.
Sanguinetti et al., 2018), annotators often fail to develop
an intersubjective understanding of what ultimately counts
as hate speech, and thus end up classifying the same re-
marks differently, depending on their own background and
personal views (Saleem et al., 2017; Salminen et al., 2018).
The common denominator of the existing NLP literature on
the matter seems to be that annotating for hate speech is in
itself particularly challenging. The most pertinent reason
for this seems to be the notorious elusiveness of the label
‘hate speech’ itself. Despite having been established in le-
gal discourse to refer to speech that can be taken to incite to
discriminatory hatred, hate speech is now often used as an
umbrella label for all sorts of hateful/insulting/abusive con-
tent (Brown, 2017). This much is strikingly evident when
one takes into account that most of the studies reviewed in
NLP state-of-the-art reports on hate speech detection (in-
cluding our discussion so far) formulate the question at
hand using different – albeit interrelated – terms, which
apart from hate speech variously include harmful speech,
offensive and abusive language, verbal attacks, hostility or
even cyber-bullying, among others. In this vein, as Waseem
et al. (2017, p.78) observe and exemplify, this “lack of con-
sensus has resulted in contradictory annotation guidelines –
some messages considered as hate speech by Waseem and
Hovy (2016) are only considered derogatory and offensive
by Nobata et al. (2016) and Davidson et al. (2017).”
The apparent confusion as to how to adequately define hate
speech is of course not exclusive to NLP research, but ex-
tends to social scientific (Gagliardone et al., 2015) and even
legal treatments of the theme (Bleich, 2014; Sellars, 2016).
Given this general confusion, it is certainly no surprise
that annotators, especially ones who do not have domain-
specific knowledge on the matter, will be prone to disagree-
ing as to how to classify some text in the relevant task, as
opposed to other comparable annotation tasks.
Discussing this issue within the more general context of de-
tecting abusive language, Waseem et al. (2017) identify
two sources for the resulting annotation confusion: (a) the
existence of abusive language directed towards some gen-
eralised outgroup, as opposed to a specific individual; and
(b) the implicitness with which an abusive attitude can of-
ten be communicated. Despite targeting abusive language
in general, the resulting two-fold typology has been taken
to apply to the more particular discussion of hate speech
too (ElSherief et al., 2018; MacAvaney et al., 2019; Mulki
et al., 2019; Rizos et al., 2019), rendering the lack of an
explicit insult/threat towards an individual target in terms
of their membership to a protected group more difficult to
classify as hate speech than a remark that explicitly incites
to discriminatory hatred. This much seems to be further
corroborated by a recent study by Salminen et al. (2019)
which revealed that, when evaluating the hatefulness of on-
line comments on a scale of 1 (not hateful at all) to 4 (very
hateful), annotators agree more on the two extremes than in
the middle ground. Quite justifiably, this suggests that it is
easier to classify directly threatening or insulting messages
as hate speech, rather than indirectly disparaging ones.
What transpires from a review of the literature then is that
the difficulty in annotating for hate speech lies primarily in
those instances of hate speech that appear to fall under the

radar of some annotators due to the ways in which incite-
ment to discriminatory hatred can be concealed in linguistic
expression. For us, this is precisely the point where CDA
can offer valuable insight towards developing schemes for
hate speech annotation. That is because CDA specifically
seeks, through fine-grained analysis and a close reading of
the text under investigation, to uncover the “consciousness
of belief and value which are encoded in the language – and
which are below the threshold of notice for anyone who ac-
cepts the discourse as ‘natural’” (Fowler, 1991, p. 67). As
we will now turn to show, an appreciation of hate speech
as an ideology-based phenomenon can substantially inform
NLP research in the area too.
3. Pilot dataset
For the purposes of the C.O.N.T.A.C.T. project we fo-
cused specifically on analysing homophobia and xenopho-
bia in local (that is, Maltese) user comment discourse (As-
simakopoulos and Vella Muskat, 2017; Assimakopoulos
and Vella Muskat, 2018). In this respect, the Maltese
C.O.N.T.A.C.T. corpus, which served as a pilot for the
presently proposed annotation scheme, was built and an-
notated following the common methodology established
across the C.O.N.T.A.C.T. consortium (Assimakopoulos
et al., 2017, pp. 17-20). The dataset was formed by
scraping Maltese portals for comments found underneath
news reports related to LGBTIQ+ and migrant minorities
over two time periods: April-June 2015 and December
2015-February 2016. The identification of relevant arti-
cles was facilitated by the use of the EMM Newsbrief we-
bcrawler3
, where we performed a search for articles from
Malta containing keywords pertaining to migrants and the
LGBTIQ+ community in turn. We then scraped comments
amounting to approximately 5,000 words worth of content
per keyword, equally distributed across the selected key-
words, eventually forming two subcorpora: one for migra-
tion, comprising 1130 comments (41020 words), and one
for LGBTIQ+ matters, comprising 1109 comments (40924
words).
Following the compilation of the corpus, we engaged in an-
notation which comprised two steps. In the first instance,
comments were classified in terms of their polarity, as pos-
itive or negative, depending on their underlying stance to-
wards the minority groups in question. Then, through an
in-depth reading of each comment labelled as negative, we
identified the discursive strategies that underlie the com-
munication of the negative attitude at hand. Crucially, this
deeper level of annotation included the detection not only
of linguistic forms, such as the use of derogatory vocabu-
lary and slurs, tropes, or generics, but also implicit prag-
matic functions that underlie the communication of insults,
threats, jokes and stereotyping.
4. The special nature of hate speech
Right from the beginning of the aforementioned annotation
of the pilot dataset, one issue that became immediately evi-
dent was that, much like in the majority of the correspond-
ing NLP research, a shallow binary classification of positive
3
http://emm.newsbrief.eu
and negative attitude cannot possibly do justice to the par-
ticularities of hate speech. This realisation appears to be in
line with the argument made independently by the several
NLP researchers who have proposed, as we have seen, more
complex annotation systems than a simple ±hate speech
classification. Against this backdrop, the minute analysis
performed on each comment for C.O.N.T.A.C.T. could not
lend itself to multiple repetitions by different annotators in
order to establish agreement, but it still reveals some prin-
cipal reasons why the traditional reliance on lexical and
grammatical features for the training of a hate speech de-
tection model falls short of capturing the intricate nature of
the category in question.
4.1. Hate speech and offensive language
As common experience indicates, the internet is full
of emotional language that can be considered insulting
and disrespectful towards both individuals and collective
groups. In this setting, online hate speech might often con-
tain offensive language, but not all offensive language used
on the internet can be considered hate speech. That is be-
cause hate speech, in its specialised legal sense, is typically
tied to the specific criterion of incitement to discrimina-
tory hatred in the several jurisdictions where it is regulated,
while the discriminatory attitude needs to specifically target
a group that is legally protected.
Indeed, since our pilot corpus was annotated for hate speech
specifically targeting migrants and the LQBTIQ+ commu-
nity, it became obvious that there were a number of com-
ments aimed at other groups of people as well as individ-
uals too. As the following response from one commenter
to another makes clear, a simple positive-negative attitude
classification could easily lead to the inaccurate labelling of
data within the corpus:
(1) I’m not your Hun pervert!
While this comment is not targeted at a minority, it is a
very direct insult, unlike much of the discourse that pointed
to minorities in our dataset, which was more often than not
only indirectly offensive. Similarly, although less abusive,
the following comment also offends another online user:
(2) I hope you’re just trolling [username]. If not, you truly
are a sad being.
Clearly, despite being obviously offensive, the two exam-
ples do not constitute hate speech, since they are not tar-
geted toward any group, much less a protected minority.
In a similar vein, both (3) and (4) do not take issue with a
minority group, but rather target groups that can be seen as
dominant in the Maltese context.
(3) The issue or problem is that religion and government
are not truly separate on Malta and never were or will be
in the future no matter what is written in any constitution.
(4) Just remember that they wouldn’t be working Illegaly if
there wasn’t someone willing to employ them Illegaly, and
a lot of these employers are Maltese.
Evidently, (3) criticises the fact that the church and state in
Malta are still intertwined, while (4) undermines the gen-
eral negative stance taken toward migrants by expounding

the hypocrisy of such a stance alongside the willingness of
the Maltese to hire members of this group and thus ben-
efit from cheap labour. Again, within the framework of
discriminatory discourse, although these examples display
a negative disposition against collective groups (church,
state, the Maltese), they cannot be deemed hate speech,
since the groups targeted are not protected under Maltese
Law.
4.2. Hate speech as a continuum
Beyond the examples which show that not all comments
labelled as negative constitute hate speech in the legisla-
tive sense of the word, a mere label of ‘negative’ fails to
capture the complexity of the scale on which various forms
of incitement to discriminatory hatred fall. In this respect,
the analysis of the C.O.N.T.A.C.T. corpus also revealed that
discriminatory hatred can vary from direct calls to violence,
as in (5), to indirect forms of discrimination, like the one
exemplified in (6):
(5) They could use a woman to execute them because they
believe that if they are killed by a woman they will not go to
heaven and enjoy their 77 virgins.
(6) I believe the problem is Europe wide, but I feel Malta is
harder hit simply because of it’s size. Malta simply has not
got the room to keep receiving these people.
While the commenter in (5) suggests that Muslim migrants
are executed, the one in (6) does not appear at first sight
to articulate anything directly offensive or insulting; upon
closer inspection, however, it can be taken to allude to ideas
of exclusion and in-grouping, since the use of ‘these peo-
ple’ serves to create an in-group of us (who belong here and
deserve access to resources) and them (who do not belong
here and should not be taking up our precious space).
While the two examples given above illustrate the two op-
posite ends of a spectrum, there is of course much that
lies in between. In the following comment, for example,
the commenter may acknowledge the need to respect mem-
bers of the LGBTIQ+ community, but concurrently refers
to specific members of the minority group, that is transgen-
der individuals, as ‘too complicated and abnormal:’
(7) We just need to teach our children to respect each other
whoever they be. Teaching gender diversity is too compli-
cated and abnormal for their standards. I’m afraid it would
effect their perception that lgbti is the norm instead of a mi-
nority.
So, despite emphasising the importance of promoting di-
versity, the commenter also recommends that children are
taught to differentiate between the ‘normal’ dominant iden-
tities and ‘abnormal’, and thus subordinate ones. Equiva-
lently, the commenter in [8] explicitly denies racism, while
at the same time clearly employing tactics of negative
stereotyping and categorisation with the use of the term ‘il-
legals.’
(8) [username] sure NO! The majority of Maltese people
are against ” ILLEGALS”! Do not mix racism with illegals
please!
All in all, as the examples above show, there are varying
degrees of discriminatory hatred that need to be accommo-
dated within the purview of hate speech. Crucially, this is
something that is underlined in non-NLP research on hate
speech too (Vollhard, 2007; Assimakopoulos et al., 2017).
Cortese (2007, p. 8-9), for example, describes a framework
that treats hate speech not as a single category, but rather as
falling on a four-point scale:
1. unintentional discrimination: unknowingly and unin-
tentionally offending a (member of a) minority group
by, for example, referring to the group by means of a
word that is considered offensive or outdated, such as
referring to black people as ‘coloured’ or by referring
to asylum seekers as ‘immigrants;’
2. conscious discrimination: intentionally and con-
sciously insulting a (member of a) minority group by,
for example, using a pejorative term like ‘faggot’ or
by describing to migrants as ‘invaders;’
3. inciting discriminatory hatred: intentionally and con-
sciously generating feelings of hatred toward minori-
ties by publicly encouraging society to hate and ex-
clude the group in question, such as by suggesting that
members of the LGBTIQ+ community are sick or that
migrants bring contagious diseases from their coun-
tries;
4. inciting discriminatory violence: intentionally and
consciously encouraging violence against minorities
by, for example, suggesting that members of a minor-
ity group be executed.
The importance of a micro-classification of our negative
comment data becomes apparent against the backdrop of
Cortese’s categorisation, since not all the examples given
above would fall within the third or fourth regions of the
scale, which are the points that most current hate speech
legislation appears to regulate. That said, given the wider
use of the term hate speech in everyday discourse, sev-
eral annotators could take texts that fall under the first two
points of Cortese’s scale to constitute hate speech too.
4.3. Hate speech and implicitness
The discussion so far inevitably leads to what we consider
to be the main challenge in achieving an intersubjective un-
derstanding of hate speech. Clearly, explicit incitement, of
the type expressed by an utterance of “kill all [members of
a minority group]” or “let’s make sure that all [members
of a minority group] do not feel welcome,” is not very dif-
ficult to discern. However, since “most contemporary so-
cieties do disapprove of” such invocations, “overt prejudi-
cial bias has been transformed into subtle and increasingly
covert expressions” (Leets, 2003, p.146). Therefore, while
clearly indicative of a negative attitude, the use of deroga-
tory terms, as in (9), cannot account for all instances of hate
speech:
(9) That’s because we’re not simply importing destitute
people. We’re importing a discredited, disheveled and de-
structive culture.

In this respect, it is often acknowledged (Warner and
Hirschberg, 2012; Gao and Huang, 2017; Schmidt and
Wiegand, 2017; Fortuna and Nunes, 2018; Sanguinetti et
al., 2018; Watanabe et al., 2018) that discourse context
plays a crucial role in evaluating remarks, as there are sev-
eral indirect strategies for expressing discriminatory ha-
tred, which our in-depth analysis enabled us to addition-
ally unearth. The most pertinent example of this is the
use of metaphor, which has for long been emphasised as
a popular strategy for communicating discrimination, since
it typically involves “mappings from a conceptual ‘source
domain’ to a ‘target domain’ with resulting conceptual
‘blends’ that help to shape popular world-views in terms of
how experiences are categorized and understood” (Musolff,
2007, p.24). In (10), for example, the commenter engages
in incitement by making use of the metaphorical schema
(Lakoff and Johnson, 2003) MIGRATION IS A DISEASE:
(10) Illegal immigration is a cancer which if not eliminated
will bring the downfall of Europe and European culture.
Similar indirect strategies can be found in the frequent use
of figurative language to underline the urgency of the sit-
uation, as exemplified by the use of allusion in (11) and
hyperbole in (12):
(11) Will Malta eventually become the New Caliphate?
(12) . . . in 4 more days we will become the minority . . .
Alongside figurative language, stereotypical representa-
tions and remarks generalising over a minority group, as
in (12) also play a crucial role in the expression of discrim-
inatory hatred (Brown, 2010; Haas, 2012; Maitra and Mc-
Gowan, 2012; Kopytowska and Baider, 2017):
(13) If anyone is lacking, it is you guys for lacking a sense
of decency. Just look at the costumes worn at gay parades
to prove my point.
Now, while such remarks might not appear to fall under
the incitement criterion for the delineation of hate speech,
there is still reason to include them in a corpus of hate
speech. As Fortuna and Nunes (2018, p.85:5) argue, “all
subtle forms of discrimination, even jokes, must be marked
as hate speech” because even seemingly harmless jokes in-
dicate “relations between the groups of the jokers and the
groups targeted by the jokes, racial relations, and stereo-
types” (Kuipers and van der Ent, 2016) and their repetition
“can become a way of reinforcing racist attitudes” (Kom-
patsiaris, 2017) and have ”negative psychological effects
for some people” (Douglass et al., 2016).
Be that as it may, people are bound to disagree as to whether
such implicit expressions can or should indeed be classi-
fied as hate speech. In view of this, we think that a strategy
for obtaining more reliable annotation results would be to
refrain as much as possible from asking raters to make a
judgement that could be influenced by their subjective opin-
ion as to what ultimately constitutes hate speech, as well
as by their own views and attitudes towards the minority
groups under question.
5. Towards a new annotation scheme
We trust that our discussion so far provides insight as to
why it is particularly difficult to achieve adequate agree-
ment among different crowd coders when it comes to clas-
sifying hate speech. Given that non-experts cannot always
distinguish between hate speech and offensive language,
and given the varying thresholds that hate speech laws place
on the continuum of discriminatory discourse, it would
probably be hard even for hate speech experts to fully agree
on a classification (Waseem, 2016; Ross et al., 2016). It is
thus clear that a simple binary classification of online posts
into hate speech or non-hate speech is unlikely to be reli-
able, and that a more complex scheme is inevitably needed.
When we presented previous annotation frameworks above,
we mentioned that the one developed by Sanguinetti et al.
(2018) was informed by critical discussions of the concept
of hate speech and was thus relatively complex. While on
the right track, however, it seems to also incorporate cate-
gories, such as intensity, aggressiveness and offensiveness,
that are susceptible to a rather subjective evaluation. In-
dispensable though they may be for the discussion of hate
speech, such categories can easily compromise agreement
as well, since individuals with different backgrounds and
views could provide markedly different interpretations of
the same data.
The proposal that we wish to make in this paper is that such
subjective categories are left out from annotation instruc-
tions as much as possible. To this effect, the scheme that
we developed for hate speech annotation in the MaNeCo
corpus is hierarchical in nature and expected to lead to the
identification of discriminatory comments along the afore-
mentioned scale by Cortese without focusing on such im-
pressionistic categories as intensity or degree of hateful-
ness, as follows:
1. Does the post communicate a positive, negative or
neutral attitude? [Positive / Negative / Neutral]
2. If negative, who does this attitude target? [Individual
/ Group]
(a) If it targets an individual, does it do so because of
the individual’s affiliation to a group? [Yes / No]
If yes, name the group.
(b) If it targets a group, name the group.
3. How is the attitude expressed in relation to the target
group? Select all that apply. [ Derogatory term /
Generalisation / Insult / Sarcasm (including jokes
and trolling) / Stereotyping / Suggestion / Threat ]4
4
The selection of these particular communicative strategies as
opposed to alternative ones is based on the categories used for
the purposes of the C.O.N.T.A.C.T. project. We thus assume an
adequate coverage, since all these strategies were identified in the
parallel annotation of corpora from the nine different national con-
texts represented in the project. That being said, we considered
adding a field for annotators to suggest their own categories, but
decided against it, since this could easily lead to confusion when
it comes to grouping different categories together during the pro-
cessing of responses.

(a) If the post involves a suggestion, is it a suggestion
that calls for violence against the target group?
[Yes / No]
With regards to the proposed annotation scheme, we cannot
of course fail to acknowledge that the first label (positive,
negative, neutral) is to a certain extent subjective. Still, we
believe that it is formulated in a way that does not require
the annotator to make a judgement in relation to a post’s
hatefulness or aggressiveness, but merely to evaluate the
commenter’s attitude in writing the comment. Obviously,
this would inevitably lead to some disagreement, but it ul-
timately calls for a judgement that should be more straight-
forward to make, as it is not tied to an individual’s under-
standing of what constitutes discrimination per se; after all,
a negative attitude can easily crop up in various other set-
tings too, like, for example, when expressing disagreement
to someone else’s post in the same thread.
So, once negative attitude is established, the second step in
the process would help to indirectly assess whether the atti-
tude can be taken to be discriminatory too, insofar as it tar-
gets a group or an individual on the basis of group member-
ship. In this vein, depending on whether the group that each
discriminatory post targets is protected by law – like mi-
grant or LGBTIQ+ individuals usually are, but politicians
or church officials are not – we can distinguish between
hate speech and merely hateful discourse.
Finally, the third and last step will help determine the posi-
tioning of a post that fulfils the penultimate and final cri-
teria in Cortese’s scale. In this regard, posts that com-
prise a suggestion would fall under incitement to discrim-
inatory hatred, while posts specifically calling for violent
actions towards the target group and those including threats
would belong to the category of incitement to discrimi-
natory violence. By the same token, posts containing in-
sults, derogatory terms, stereotyping, generalisations and
sarcasm would fall under Cortese’s categories of conscious
and unintentional discrimination, which are not – at face
value – regulated by hate speech law, but are still, as we
have seen, relevant to the task at hand. In this way, one
should be able to establish different sets of discriminatory
talk, which could then be included or excluded from subse-
quent analyses, depending on the threshold that one speci-
fies for the delineation of hate speech.
Apart from enabling us to distinguish between explicit hate
speech and softer forms of discrimination, a further merit of
the proposed annotation scheme is that it indirectly allows
us to control for annotator disagreement in some cases. The
most obvious such case would be the presence of posts that
are ambiguous between a literal and a sarcastic interpreta-
tion, or even the identification of lexical items and general-
isations that might be considered offensive by some anno-
tators and not by others. At the same time, it could provide
useful indications regarding the distinction between con-
scious and unintentional discrimination. The rule of the
thumb here would be that the more annotators agree on a
comment belonging to the first two categories of discrim-
ination in Cortese’s scale, in the sense that it is discrimi-
natory but does not include a suggestion regarding the tar-
get group, the closer that comment would be to conscious
discrimination. That is because, as we have seen, uninten-
tional discrimination has a marked tendency to go unno-
ticed and is thus expected to be less noticed by the annota-
tors. Crude though this generalisation might be, we believe
that it still provides a criterion that is viable.5
Obviously, this is not to say that disagreement can be com-
pletely eradicated. As a matter of fact, we cannot em-
phasise enough that the aim of our suggested annotation
scheme (or of any other such scheme for that matter) is not
to achieve perfect agreement. After all, as our preliminary
CDA analysis revealed, genuine ambiguity can present it-
self not only at the structural, but also at the attitudinal
level. For example, the following comment was posted in
reaction to an article about a Nigerian rape victim who was
fighting to bring one of her two children to Malta:
(14) For God’s sake, make the process simpler and allow
this woman to unite with her daughter in Nigeria.
At face value, it seems difficult to discern if the commenter
wishes for the woman in question, who is stranded in Malta
due to not having received refugee statues, to be allowed
to reunite with her daughter by bringing her to Malta, or is
suggesting that she be deported back to Nigeria in order to
be with her daughter. Apart from this, some posts may still
be difficult to classify on the ground that they express mul-
tiple attitudes. For example, in stating (15), the commenter
acknowledges that marriage should be a matter of choice
between two people in love, but then directly implies that
marriage should be a union between two people of the op-
posite (cis)gender.
(15) People marry because they fall in love, and although
it’s a choice, it was meant to be like that even in the animal
kingdom, for example swans mate for life, male and female,
not male and male.
Despite the inevitable presence of such borderline am-
biguous comments, however, we do expect our annotation
scheme to fare better in terms of inter-annotator agreement
than previous attempts, and particularly attempts based on
a binary ±hate speech classification. In an attempt to assess
its efficacy then, we conducted a preliminary pilot study to
which we will now turn.
5.1. Piloting the new scheme
A total of 24 annotators took part in this pilot study.
The participants, who were mostly academics and students
ranging between 21 and 60 years of age, were divided into
two gender- and age-balanced groups of 12. The first group
was asked to label items using simple binary annotation
(±hate speech) on the basis of the definition of hate speech
provided within the Maltese Criminal Code, while the sec-
ond used our proposed multi-level annotation scheme. Both
groups were presented with 15 user-generated comments
from the Maltese C.O.N.T.A.C.T. corpus in random order.
5
An alternative here would potentially involve some kind of
user profiling that would allow for an identification of the com-
menter’s more general stance towards the target minority. Such
an alternative, however, could compromise the overall task, since
grouping a user’s comments together is bound to bias annotators.

In an effort to ensure variation of the items to be anno-
tated, we went back to our original CDA classification and
selected three comments from each of the following cate-
gories:
• comments involving incitement to discriminatory vio-
lence against the migrant minority;
• comments that were labeled as discriminatory towards
the migrant minority but not as fulfilling the incite-
ment criterion;
• comments that were labeled as negative but do not tar-
get the migrant minority;
• comments that were labeled as expressing a positive
attitude towards the migrant minority; and
• comments that were labeled as ambiguous, along sim-
ilar lines to the discussion of (14) and (15) above.
All individuals in both annotation conditions performed the
task independently. Obviously, in order to meaningfully
compare the levels of inter-annotator agreement for the two
schemes, we needed to have an equal number of classes
across the board. To achieve this we screened the annota-
tions received for the multi-level condition, with a view to
inferring a binary ±hate speech classification in this case
too.
While inferring a binary class from the multi-level scheme
might seem counter-intuitive, in view of the preceding dis-
cussion, there are several reasons why we did this. First,
the decision was taken on pragmatic grounds: in order to
achieve a fair comparison, agreement within the two groups
needed to be estimated based on similar categories. That
way, we were able to compute inter-annotator agreement
between participants in the two annotation tasks in a way
that would enable a direct comparison. More importantly,
however, there are theoretical grounds for using the multi-
level scheme to achieve a binary classification. One of
the arguments in the previous section was that hate speech
would be evident in comments that would be labeled as neg-
ative, targeting (members of) a minority group and com-
prising either a suggestion or a threat. Indeed, the thrust of
the argument presented above is not that a binary classifica-
tion is undesirable, but that in order to be reliably made, it
had to supervene on micro-decisions that took into account
these various dimensions. The extent to which this is the
case is of course an empirical question, one to which we
return in the concluding section.
In Table 5.1., we report percent agreement, Fleiss’s kappa
(Fleiss, 1971), and Randolph’s kappa (Randolph, 2005),
which is known to be more suitable when raters do not
have prior knowledge of the expected distribution of the
annotation categories. All in all, the results from this pilot
appear to corroborate our prediction: the proposed multi-
level annotation scheme appears to indeed lead to higher
inter-annotator agreement, which for Fleiss’s kappa in par-
ticular seems to even mark a rise from moderate (0.54) to
substantial agreement (0.69).
Apart from enabling us to indirectly compare its perfor-
mance in relation to the corresponding binary classification
Metric binary multi-level
Percent agr. 76.8% 84.6%
Fleiss’ k 0.54 0.69
Randolph’s k 0.48 0.58
Table 1: Inter-annotator agreement for the two annotation
schemes
scheme, the proposed multi-level scheme also allowed us to
pinpoint those levels at which disagreement was more pro-
nounced. So, whereas participants agreed more when asked
about the positive/negative/neutral attitude of the speaker,
as well as about the individual/group targeted by a com-
ment, they agreed far less when asked to evaluate more con-
cretely how a negative attitude is expressed in each com-
ment. Informal discussions with the participants upon com-
pletion of the task suggested that this was due to their lack
of experience with discerning the relevant discourse strate-
gies, but additionally shed light on one further source of
disagreement: some annotators expressed difficulty in dis-
tancing themselves from the attitude of the commenter, and
felt tempted to rate a comment as negative if the commenter
expressed an attitude opposing their own. Clearly, this is
valuable feedback about where we would need to focus our
attention while further improving the annotation scheme
and accompanying instruction text.
6. Future steps: The MaNeCo corpus
Having motivated the proposed annotation framework on
the basis of our pilot study, we are currently in the process
of implementing it in the annotation of various samples for
the MaNeCo corpus. At the moment, the MaNeCo cor-
pus comprises original data donated to us by the Times of
Malta, the newspaper with the highest circulation in Malta.
More specifically, it contains – in anonymised form – all
the comments from the inception of the Times of Malta on-
line platform (April 2008) up to January 2017 (when the
data was obtained). This amounts to over 2.5 million user
comments (over 124 million words), which are written in
English, Maltese or a mixture of the two languages, even
though the newspaper itself is English-language. Our aim
is to eventually populate it with data from other local news
outlets too to ensure an equal representation of opinions ir-
respective of a single news portal’s political and ideological
affiliations. That said, even this dataset on its own is an in-
valuable resource for hate speech annotation, since it also
includes around 380K comments that have been deleted
by the newspaper moderators, and which are obviously the
ones that tend to be particularly vitriolic. In this regard, see-
ing that there are generally “much fewer hateful than benign
comments present in randomly sampled data, and therefore
a large number of comments have to be annotated to find
a considerable number of hate speech instances” (Schmidt
and Wiegand, 2017, p.7), MaNeCo is well suited for ad-
dressing the challenge of building a set that is balanced be-
tween hate speech and non-hate speech data.
Clearly, this does not mean that we will not have addi-
tional challenges to face before we end up with a dataset
that could potentially be used for training and testing pur-

poses. For one, being a corpus of user-generated content in
the bilingual setting of Malta, it contains extensive code-
switching and code-mixing (Rosner and Farrugia, 2007;
Elfardy and Diab, 2012; Sadat et al., 2014; Eskander et
al., 2014), which is further complicated by the inconsistent
use of Maltese spelling online, particularly in relation to
the specific Maltese graphemes [ċ], [ġ], [gh̄], [h̄], and [ż],
for which diacritics are often omitted in online discourse.
Then, given the casual nature of most communication on
social media, it is full of non-canonical written text (Bald-
win et al., 2013; Eisenstein, 2013) exhibiting unconven-
tional orthography, use of arbitrary abbreviations and so on.
Even so, these are challenges that can be faced after the ef-
fectiveness of our proposed scheme in yielding better inter-
annotator agreement results is established more concretely.
In this regard, a line of future work, briefly discussed in
Section 3., is to further validate the multi-level annotation
schemes. Specifically, reliability needs to be estimated on
the basis of larger and more diverse samples. Furthermore,
we have already identified the question of whether, having
conducted a micro-analysis of user-generated texts, it be-
comes easier to classify them, in a second step, as instances
of hate or non-hate speech, thereby going from a multi-level
to a binary classification. The results from our pilot study
are encouraging, in that they evince higher agreement when
annotators have their attention drawn to different aspects of
the text in question. Whether this will also translate into a
more reliable detection rate for hate speech – by humans or
by classification algorithms, for which the multi-level an-
notation may provide additional features – is an empirical
question we still need to test.
In closing, to the extent that we can assess our proposed
annotation scheme’s usefulness for training automatic clas-
sifiers, we believe that, by distinguishing between differ-
ent target groups from the beginning, this scheme could
prospectively enable us to select training material from dif-
ferent domains of hate speech (such as racism, sexism,
homophobia, etc.) in a way that serves transfer learning
too. This could be particularly useful in addressing the
notorious challenge of retaining a model’s good perfor-
mance on a particular dataset source and/or domain of hate
speech during transfer learning to other datasets and do-
mains (Agrawal and Awekar, 2018; Arango et al., 2019).
After all, as Gröndahl et al. (2018) concluded, after demon-
strating – through replication and cross-application of five
model architectures – that any model can obtain compara-
ble results across sources/domains insofar as it is trained on
annotated data from within the same source/domain, future
work in automatic hate speech detection “should focus on
the datasets instead of the models,” and more specifically on
a comparison of “the linguistic features indicative of differ-
ent kinds of hate speech (racism, sexism, personal attacks
etc.), and the differences between hateful and merely of-
fensive speech.” (Gröndahl et al., 2018, p.11). This is a di-
rection that we are particularly interested in, since our CDA
research on the Maltese C.O.N.T.A.C.T dataset correspond-
ingly revealed that although users discussing the LGBTIQ+
community and migrants appear to employ similar tactics
in expressing discriminatory attitudes, the content of their
utterances differs to a considerable extent depending on the
target of their comments. 6
7. Acknowledgements
The research reported in this paper has been substantially
informed by the original work conducted by the Univer-
sity of Malta team on the C.O.N.T.A.C.T. Project, which
was co-funded by the Rights, Equality and Citizenship Pro-
gramme of the European Commission Directorate-General
for Justice and Consumers (JUST/2014/RRAC/AG). We
gratefully acknowledge the support of the Times of Malta in
making the MaNeCo data available to us. Last but not least,
we would like to thank Slavomir Ceplo for his diligent work
in extracting and organising the MaNeCo corpus, the three
anonymous LREC reviewers for their insightful comments
and, of course, all participants of the pilot study for their
generous availability.
8. Bibliographical References
Agrawal, S. and Awekar, A. (2018). Deep learning for de-
tecting cyberbullying across multiple social media plat-
forms. In Gabriella Pasi, et al., editors, Advances in In-
formation Retrieval, pages 141–153, Cham. Springer.
Arango, A., Pérez, J., and Poblete, B. (2019). Hate speech
detection is not as easy as you may think: A closer look
at model validation. In Proceedings of the 42nd Inter-
national ACM SIGIR Conference on Research and De-
velopment in Information Retrieval, pages 45–54, New
York. ACM.
Assimakopoulos, S. and Vella Muskat, R. (2017). Ex-
ploring xenophobic and homophobic attitudes in Malta:
Linking the perception of social practice with textual
analysis. Lodz Papers in Pragmatics, 13(2):179–202.
Assimakopoulos, S. and Vella Muskat, R. (2018). Xeno-
phobic and homophobic attitudes in online news portal
comments in Malta. Xjenza Online, 6(1):25–40.
Assimakopoulos, S., Baider, F. H., and Millar, S. (2017).
Online Hate Speech in the European Union: A
Discourse-Analytic Perspective. Springer, Cham.
Badjatiya, P., Gupta, S., Gupta, M., and Varma, V. (2017).
Deep learning for hate speech detection in tweets. In
Proceedings of the 26th International Conference on
World Wide Web Companion, pages 759–760, Geneva.
International World Wide Web Conferences.
Baldwin, T., Cook, P., Lui, M., MacKinlay, A., and Wang,
L. (2013). How noisy social media text, how diffrnt so-
cial media sources? In Proceedings of the Sixth Interna-
tional Joint Conference on Natural Language Process-
ing, pages 356–364, Nagoya. Asian Federation of Natu-
ral Language Processing.
Banks, J. (2010). Regulating hate speech online. In-
ternational Review of Law, Computers & Technology,
24(3):233–239.
Bleich, E. (2014). Freedom of expression versus racist hate
speech: Explaining differences between high court reg-
6
For example, while migrants are often referred to as ‘ungrate-
ful’ or ‘dangerous’ and migration is typically framed in terms of
metaphors of invasion, members of the LGBTIQ+ community are
characterised as ‘abnormal’ and ‘sinners’ with frequent appeal to
metaphors of biblical doom.

ulations in the usa and europe. Journal of Ethnic and
Migration Studies, 40(2):283–300.
Bretschneider, U. and Peters, R. (2017). Detecting Offen-
sive Statements towards Foreigners in Social Media. In
Proceedings of the 50th Hawaii International Confer-
ence on System Sciences, Maui. HICSS.
Brown, R. (2010). Prejudice: Its Social Psychology. Wi-
ley, Oxford.
Brown, A. (2017). What is hate speech? Part 2: Family
resemblances. Law and Philosophy, 36(5):561–613.
Burnap, P. and Williams, M. L. (2015). Cyber hate speech
on twitter: An application of machine classification and
statistical modeling for policy and decision making. Pol-
icy & Internet, 7(2):223–242.
Davidson, T., Warmsley, D., Macy, M., and Weber,
I. (2017). Automated hate speech detection and the
problem of offensive language. In Proceedings of the
Eleventh International AAAI Conference on Web and So-
cial Media, Palo Alto. AAAI Press.
de Gibert, O., Perez, N., Garcı́a-Pablos, A., and Cuadros,
M. (2018). Hate speech dataset from a white supremacy
forum. In Proceedings of the 2nd Workshop on Abusive
Language Online (ALW2), pages 11–20, Brussels. Asso-
ciation for Computational Linguistics.
Djuric, N., Zhou, J., Morris, R., Grbovic, M., Radosavlje-
vic, V., and Bhamidipati, N. (2015). Hate speech de-
tection with comment embeddings. In Proceedings of
the 24th International Conference on World Wide Web,
pages 29–30, New York. ACM.
Douglass, S., Mirpuri, S., English, D., and Yip, T. (2016).
“they were just making jokes”: Ethnic/racial teasing and
discrimination among adolescents. Cultural Diversity
and Ethnic Minority Psychology, 22(1):69–82.
Eisenstein, J. (2013). What to do about bad language on
the internet. In Proceedings of the 2013 Conference of
the North American Chapter of the Association for Com-
putational Linguistics: Human Language Technologies,
pages 359–369, Atlanta. Association for Computational
Linguistics.
Elfardy, H. and Diab, M. (2012). Token level identification
of linguistic code switching. In Proceedings of COLING
2012: Posters, pages 287–296, Mumbai. COLING 2012
Organizing Committee.
ElSherief, M., Kulkarni, V., Nguyen, D., Wang, W., and
Belding, E. (2018). Hate lingo: A target-based linguistic
analysis of hate speech in social media. In Proceedings
of the 12th International AAAI Conference on Web and
Social Media, Palo Alto. AAAI Press.
Eskander, R., Al-Badrashiny, M., Habash, N., and Ram-
bow, O. (2014). Foreign words and the automatic pro-
cessing of Arabic social media text written in Roman
script. In Proceedings of the First Workshop on Com-
putational Approaches to Code Switching, pages 1–12,
Doha. Association for Computational Linguistics.
Fleiss, J. L. (1971). Measuring nominal scale agreement
among many raters. Psychological Bulletin, 76:378–38.
Fortuna, P. and Nunes, S. (2018). A survey on automatic
detection of hate speech in text. ACM Comput. Surv.,
51(4):85:1–85:30.
Gagliardone, I., Gal, D., Alves, T., and Martinez, G.
(2015). Countering online hate speech. Unesco Publish-
ing, Paris.
Gao, L. and Huang, R. (2017). Detecting online hate
speech using context aware models. In Proceedings of
the International Conference Recent Advances in Natu-
ral Language Processing, RANLP 2017, pages 260–266,
Varna. INCOMA.
Greevy, E. and Smeaton, A. F. (2004). Classifying racist
texts using a support vector machine. In Proceedings of
the 27th Annual International ACM SIGIR Conference
on Research and Development in Information Retrieval,
SIGIR ’04, pages 468–469, New York. ACM.
Gröndahl, T., Pajola, L., Juuti, M., Conti, M., and Asokan,
N. (2018). All you need is ”love”: Evading hate speech
detection. In Proceedings of the 11th ACM Workshop
on Artificial Intelligence and Security, pages 2–12, New
York. ACM.
Haas, J. (2012). Hate speech and stereotypic talk. In
H. Giles, editor, The Handbook of Intergroup Commu-
nication, pages 128–140. Routledge, London.
Kompatsiaris, P. (2017). Whitewashing the nation: racist
jokes and the construction of the african ‘other’ in greek
popular cinema. Social Identities, 23(3):360–375.
Kopytowska, M. and Baider, F. (2017). From stereo-
types and prejudice to verbal and physical violence:
Hate speech in context. Lodz Papers in Pragmatics,
13(2):133–152.
Kuipers, G. and van der Ent, B. (2016). The seriousness
of ethnic jokes: Ethnic humor and social change in the
netherlands, 1995–2012. HUMOR, 29(4):605–633.
Kwok, I. and Wang, Y. (2013). Locate the hate: De-
tecting tweets against blacks. In Proceedings of the
Twenty-Seventh AAAI Conference on Artificial Intelli-
gence, pages 1621–1622, Palo Alto. AAAI Press.
Lakoff, G. and Johnson, M. (2003). Metaphors we live by.
University of Chicago Press, Chicago.
Leets, L. (2003). Disentangling perceptions of subtle racist
speech: A cultural perspective. Journal of Language and
Social Psychology, 22(2):145–168.
MacAvaney, S., Yao, H., Yang, E., Russell, K., Goharian,
N., and Frieder, O. (2019). Hate speech detection: Chal-
lenges and solutions. PloS One, 14(8):1–16.
Maitra, I. and McGowan, M. K. (2012). Introduction and
overview. In I. Maitra et al., editors, Speech and harm:
Controversies over free speech, pages 1–23. Oxford Uni-
versity Press, Oxford.
Malmasi, S. and Zampieri, M. (2018). Challenges in
discriminating profanity from hate speech. Journal
of Experimental & Theoretical Artificial Intelligence,
30(2):187–202.
Mulki, H., Haddad, H., Bechikh Ali, C., and Alshabani,
H. (2019). L-HSAB: A Levantine twitter dataset for
hate speech and abusive language. In Proceedings of
the Third Workshop on Abusive Language Online, pages
111–118, Florence. Association for Computational Lin-
guistics.
Musolff, A. (2007). What role do metaphors play in racial

prejudice? the function of antisemitic imagery in hitler’s
mein kampf. Patterns of Prejudice, 41(1):21–43.
Nobata, C., Tetreault, J., Thomas, A., Mehdad, Y., and
Chang, Y. (2016). Abusive language detection in on-
line user content. In Proceedings of the 25th Interna-
tional Conference on World Wide Web, WWW ’16, pages
145–153, Geneva. International World Wide Web Con-
ferences.
Randolph, J. J. (2005). Free-marginal multirater kappa:
An alternative to Fleiss’ fixed-marginal multirater kappa.
Rizos, G., Hemker, K., and Schuller, B. (2019). Aug-
ment to Prevent: Short-Text Data Augmentation in Deep
Learning for Hate-Speech Classification. In Proceedings
of the 28th ACM International Conference on Informa-
tion and Knowledge Management - CIKM ’19, pages
991–1000, Beijing. ACM.
Rosner, M. and Farrugia, P.-J. (2007). A tagging algorithm
for mixed language identification in a noisy domain. In
INTERSPEECH 2007: 8th Annual Conference of the In-
ternational Speech Communication Association, pages
190–193, Red Hook. Curran Associates.
Ross, B., Rist, M., Carbonell, G., Cabrera, B., Kurowsky,
N., and Wojatzki, M. (2016). Measuring the Reliability
of Hate Speech Annotations: The Case of the European
Refugee Crisis. In Michael Beißwenger, et al., editors,
Proceedings of NLP4CMC III: 3rd Workshop on Natural
Language Processing for Computer-Mediated Commu-
nication, pages 6–9, Frankfurt. Bochumer Linguistische
Arbeitsberichte.
Sadat, F., Kazemi, F., and Farzindar, A. (2014). Automatic
identification of Arabic language varieties and dialects
in social media. In Proceedings of the Second Workshop
on Natural Language Processing for Social Media (So-
cialNLP), pages 22–27, Dublin. Association for Compu-
tational Linguistics and Dublin City University.
Saleem, H. M., Dillon, K. P., Benesch, S., and Ruths, D.
(2017). A web of hate: Tackling hateful speech in on-
line social spaces. In Proceedings of the First Workshop
on Text Analytics for Cybersecurity and Online Safety
at LREC 2016, Portorož. European Language Resources
Association (ELRA).
Salminen, J., Veronesi, F., Almerekhi, H., Jung, S., and
Jansen, B. J. (2018). Online hate interpretation varies
by country, but more by individual: A statistical analysis
using crowdsourced ratings. In Proceedings of the Fifth
International Conference on Social Networks Analysis,
Management and Security (SNAMS), pages 88–94, Red
Hook. Curran Associates.
Salminen, J., Almerekhi, H., Kamel, A. M., Jung, S.-g.,
and Jansen, B. J. (2019). Online hate ratings vary by
extremes: A statistical analysis. In Proceedings of the
2019 Conference on Human Information Interaction and
Retrieval, CHIIR ’19, pages 213–217, New York. ACM.
Sanguinetti, M., Poletto, F., Bosco, C., Patti, V., and
Stranisci, M. (2018). An Italian twitter corpus of
hate speech against immigrants. In Proceedings of the
Eleventh International Conference on Language Re-
sources and Evaluation (LREC 2018), Portorož. Euro-
pean Language Resources Association (ELRA).
Schmidt, A. and Wiegand, M. (2017). A survey on hate
speech detection using natural language processing. In
Proceedings of the Fifth International Workshop on Nat-
ural Language Processing for Social Media, pages 1–10,
Valencia. Association for Computational Linguistics.
Sellars, A. (2016). Defining hate speech. Berkman Klein
Center Research Publication No. 2016-20; Boston Uni-
versity School of Law, Public Law Research Paper No.
16-48.
Spertus, E. (1997). Smokey: Automatic recognition of
hostile messages. In AAAI-97 - Proceedings of the Four-
teenth National Conference on Artificial Intelligence and
The Ninth Annual Conference on Innovative Applica-
tions of Artificial Intelligence, pages 1058–1065, Boston.
MIT Press.
Tulkens, S., Hilte, L., Lodewyckx, E., Verhoeven, B., and
Daelemans, W. (2016). The automated detection of
racist discourse in dutch social media. Computational
Linguistics in the Netherlands Journal, 6(1):3–20.
Warner, W. and Hirschberg, J. (2012). Detecting hate
speech on the world wide web. In Proceedings of the
Second Workshop on Language in Social Media, pages
19–26, Montréal. Association for Computational Lin-
guistics.
Waseem, Z. and Hovy, D. (2016). Hateful symbols or hate-
ful people? predictive features for hate speech detection
on twitter. In Proceedings of the NAACL Student Re-
search Workshop, pages 88–93, San Diego. Association
for Computational Linguistics.
Waseem, Z., Davidson, T., Warmsley, D., and Weber, I.
(2017). Understanding abuse: A typology of abusive
language detection subtasks. In Proceedings of the First
Workshop on Abusive Language Online, pages 78–84,
Vancouver. Association for Computational Linguistics.
Waseem, Z. (2016). Are you a racist or am I seeing things?
annotator influence on hate speech detection on twitter.
In Proceedings of the First Workshop on NLP and Com-
putational Social Science, pages 138–142, Austin. Asso-
ciation for Computational Linguistics.
Watanabe, H., Bouazizi, M., and Ohtsuki, T. (2018). Hate
speech on twitter: A pragmatic approach to collect hate-
ful and offensive expressions and perform hate speech
detection. IEEE Access, 6:13825–13835.
Zampieri, M., Malmasi, S., Nakov, P., Rosenthal, S., Farra,
N., and Kumar, R. (2019). Predicting the type and target
of offensive posts in social media. In Proceedings of the
2019 Conference of the North American Chapter of the
Association for Computational Linguistics: Human Lan-
guage Technologies, Volume 1 (Long and Short Papers),
pages 1415–1420, Minneapolis. Association for Compu-
tational Linguistics.

Annotating For Hate Speech The MaNeCo Corpus And Some Input From Critical Discourse Analysis

More Related Content

Similar to Annotating For Hate Speech The MaNeCo Corpus And Some Input From Critical Discourse Analysis (20)

More from Sabrina Baloi (20)

Recently uploaded (20)

Annotating For Hate Speech The MaNeCo Corpus And Some Input From Critical Discourse Analysis