Researcher: Hate, dangerous speech low among Ethiopian facebook users

Ethiopians’ discussion on facebook is much more nuanced than perceived to be, according to a team of researchers monitoring the social media.

The lead researcher, Iginio Gagliardone, says the ratio of hate speech and dangerous speech is negligible.

“Only 0.4% of statements in our sample have been classified as hate speech and 0.3% as dangerous speech” [and] “1.5% of statements has been classified as offensive”, Iginio Gagliardone said during an exclusive interview with HornAffairs.

Yet, one out of seven statements analysed were found to be “attacking another speaker or a specific group by belittling, provoking, teasing them maliciously, or explicitly threatening them”.  That might explain why some users perceive nasty statements dominate facebook. Such statements, however, do not always qualify as hate speech or dangerous speech.

“Often political opponents accuse each other of producing hate speech, but when analysed, most statements are actually just political insults”, added Iginio Gagliardone, a lecturer at the University of the Witwatersrand in South Africa and associate research fellow at the University of Oxford.

The findings also shows that about half of the statements analysed were constructive, even if they may contain strong criticism.

The research covered 1055 facebook pages and 13,000 statements monitored between February and April last year. The project – dubbed Mechachal, an Amharic equivalent for tolerance – consisted academics from the University of Oxford and Addis Ababa University and has so far released two interim reports.

“These findings have broad implications for researchers studying online hate speech and for policy makers seeking to promote targeted responses to it. To our knowledge, this is the first time, a research focusing on an entire country, and its diaspora community, has been able to quantify the proportion of hate and dangerous speech among the total conversations occurring on a social networking platform”, Iginio Gagliardone said.

HornAffairs’ Daniel Berhane interviewed Iginio Gagliardone on the methodologies and findings of the research.

Photo - Iginio Gagliardone
Photo – Iginio Gagliardone

HornAffairs: The research project sought to analyze Ethiopia online debates, however, you actually focused on facebook. Why was that?

Iginio Gagliardone: There are various reasons for this choice. First, it is based on the relevance of Facebook for Ethiopians in terms of accessing the Internet in Ethiopia and abroad. We do not have hard evidence about the comparative importance of Facebook in relation to other online platforms, but interviews with key figures in Ethiopia’s media and political landscapes (both in government and in the opposition), emphasised how, for them, Facebook is the most important platform for reaching out to online audiences, and especially the youth.

Secondly, we wanted to analyse a space that is open to any kind of debate, rather than focusing on websites or blogs that have already an agenda. By doing so, we could capture both casual conversations and politically heated debates.

Finally, comparative studies have shown how in countries with limited Internet penetration, like Ethiopia, Facebook has become almost a synonym for the Internet, a platform through which users access information, services, and participate in online debates.

HornAffairs: Facebook is not geographically segregated, how did you manage to analyze Ethiopian discourse in particular?

Iginio Gagliardone: We were advantaged by the uniqueness of the languages spoken in Ethiopia, especially Amharic, Tigrigna, and Oromiffa. Many posts are written in English. But we did not select just posts. We initially selected Facebook spaces – be it individual profiles, pages, or open groups. If these spaces were focusing on Ethiopia, it was likely to find at least some content in one of the main languages spoken by Ethiopians in Ethiopia and abroad. Once the selection of relevant spaces was made, we then selected individual statements appearing on them, independently from the language used.

HornAffairs: How many pages and threads did you monitor?

Iginio Gagliardone: We monitored 1055 pages in total, and around 13,000 statements. Some pages were followed regularly, such as Dimstachin Yesema and Dr Tedros Adhanom. Other pages were randomly selected on a daily basis among the total 1055.

HornAffairs: I gather your focus was on hate speech.

Iginio Gagliardone: Mechachal began with an interest in identifying and analysing hate speech, especially messages with the highest likelihood of leading to violence. This spirit informed the project’s first pilot study, which was conducted between October 2013 and February 2014. The pilot was limited in scope but it highlighted at least two important challenges for research on hate speech online.

First, the need to understand not just the nature, but also the prevalence of hate speech in the context of broader online debates. The growing concerns about uses of social media that can promote radicalization and incite violent acts has led to increasing demands for research that can detect and monitor these types of online behaviours. As attention towards the “darker” side of the Internet grows, however, in the absence of data that can contextualize how significant extreme speech actually is, it may lead to creating an image of online conversations as a whole as potentially harmful. This challenge led to developing a strategy that allows not just detecting the most extreme forms of speech (as has been the case so far for most projects focusing on hate speech online), but also measuring how prevalent they are among conversations in social media.

The second challenge is related to the importance of simultaneously detecting and understanding the functioning of spaces of engagement, where users seek to create communicative relationships across divides, rather than exacerbating existing tensions. We thus sought to divide selected statements among those going against and those going towards.

HornAffairs: How do you classify statements as “going against” and “going towards”?

Iginio Gagliardone: Our classification is rooted in research and theories of deliberation, political conflict, and political engagement, and it has been progressively tailored to apply to Ethiopia’s political environment.

Going towards and going against are not about agreeing or disagreeing, but more about the tendency to take a viewpoint seriously and engage with it, or, on the contrary, to dismiss it and directly attack a person for his/her affiliation with a specific group.

Statements that go against are statements attacking another speaker or a specific group by belittling, provoking, teasing them maliciously, or explicitly threatening them. The type of ‘name-calling’ that has characterized some of the debates in Ethiopia’s traditional and new media, for example, would fall in this category. Terms such as ‘anti-peace’ and ‘chauvinist’, or puns replacing ‘prime minister’ with ‘crime minister’, have been used frequently to dismiss adversaries, removing possibilities of engaging with an issue a priori, because its source made it unworthy of attention.

Statements that go towards, on the contrary, are statements that help initiate, maintain, and/or build a communicative relationship, for example acknowledging another person or group’s position, offering additional information about the topic being discussed, joking (in a non-hostile teasing way), and creating engagement and conversation with the other members in the discussion. Statements going towards can also contain strong criticism, but in the context of a polarized debate, they offer at least a premise for recognizing adversaries as legitimate, rather than simply dismissing them.

These concepts cut the continuum of statements uttered in online debates into two broad categories and represent the beginning of more in-depth analysis. More details can be found in the two reports that we have published so far and are available online. A third, more comprehensive report is going to be published later in May.

Graph - Mechachal, Online Debates and Elections in Ethiopia
Graph – Mechachal, Online Debates and Elections in Ethiopia

HornAffairs: What’s your finding regarding Ethiopians discourse facebook? How prevalent are statements going against the speaker (i.e. attacking, dismissing the speaker)?

Iginio Gagliardone: Let me point out that, in the case of statements going against, four sub-categories have been created. That is – Offensive speech; Hate speech; Dangerous speech with limited possibility for the speakers (or the groups they appeal to) to carry out violence; and Dangerous speech with high risk that the speakers (or the groups they appeal to) could carry out violence.

The main findings illustrate that, using these framework and categories, hate and dangerous speech are marginal forms of speech in Ethiopia, at least on Facebook. Only 0.4% of statements in our sample have been classified as hate speech (i.e. speech that incites others to discriminate or act against individuals or groups based on their ethnicity, religion, or gender) and 0.3% as dangerous speech (i.e. speech that builds the bases for or directly calls for widespread violence against a particular group).

Among the statements categorized dangerous speech, all statements were associated with a limited possibility for the speakers (or the groups they appeal to) to actually carry out violence. No statements were found in our general sample to have a high risk that the speakers or the groups they appeal to could carry out violence.

HornAffairs: Many people would argue hate speech and dangerous speech are more than one percent.  What is your finding regarding offensive speech?

Iginio Gagliardone: 1.5% of statements have been classified as offensive, and 16% of statements as going against. This can account for the perception some Internet users may have of Facebook being used more extensively to attack adversaries. It is important, however, to distinguish these forms of speech, that can be considered disturbing by some, from hate speech. As fellow researchers in Kenya have also pointed out (See UMATI Project Final Results[pdf]), often political opponents accuse each other of producing hate speech, but when analysed, most statements are actually just political insults.

HornAffairs: Let me raise specific questions from your second report, which analysed two months (February 24 – March 24 and March 25 to April 24, 2015).

Iginio Gagliardone: You may.

HornAffairs: Your report claims that, “contrary to expectations, as Election Day drew nearer, antagonistic statements actually decreased”. How plausible is the conclusion?

Iginio Gagliardone: On Facebook, the 2015 election was treated largely as a “non-event”. Most Facebook pages discussed the elections, but many statements either directly referred to, or seemed informed by, the perception that the outcome of the elections was already predetermined, with low levels of suspense and low expectations on the part of online users.

Antagonistic speech was mostly directed against the elections themselves. Venting against the electoral process often took the form of sarcastic jokes and remarks, which showed disillusionment and frustration. Because the elections were seen to have a foregone outcome, little political campaigning took place on Facebook, even if opposition parties took greater advantage of the opportunities offered by social media than the EPRDF.

Opposition supporters tended to gravitate towards Facebook much more than government supporters, as shown by the higher ratio of positive feedback received by opposition-affiliated pages. Campaigning on Facebook appeared to have the primary objective of reinforcing the sense of identity of the various online communities, and was therefore more an act of community-building than a competitive political effort.

HornAffairs: You also analysed six selected facebook pages/timelines. What did it add to your general observation?

Iginio Gagliardone: There are a number of findings that emerged from the regular analyses of key online spaces. For example we noticed that, despite Facebook’s refers to people following updates from a public figure or popular page in terms of followers and fans, a complex and often adversarial relationship exists between the owners of those spaces and their “audience”. This is reflected both in the issues that are discussed – with commenters often moving away from a topic suggested in a post to address other issues – and in the tone of the discussions, whereby commentators tend to be more antagonistic than the public figures or institutions owning the page(s).

Analysing conversation of Facebook as a whole, we found that: Users with little or no influence tend to post more statements going against (18% of the total of statements analysed) than highly influential speakers (11%). The criteria we applied to define influence (and are specified in the two reports) are limited to online spaces and based on a user’s follower base. This decision was made when the team was confronted with the complexities of assessing a user’s influence offline, and in order to account for the possibility of a speaker gaining influence through their online presence and activity.

The gap between the tone of the statements between influential and non-influential speakers in the six pages we analysed is significantly higher. Only a small fraction of posts on DireTube, Dr Tewodros, or the Semayawi party go against (less than 7%), while more than 25% of the comments on the same pages go against.

HornAffairs: Anything you would like to add?

Iginio Gagliardone: Taken together, these findings seem to delineate a picture where social media are acting as a space allowing ordinary citizens to express their anger towards power, rather than as a tool used by influential people to radicalize other individuals. This does not rule out the possibility of social media being used with this particular aim, but it suggests that if we focus on broader trends, these possibilities are fairly limited in scope and scale.

These findings have broad implications for researchers studying online hate speech and for policy makers seeking to promote targeted responses to it. To our knowledge this is the first time, a research focusing on an entire country, and its diaspora community, has been able to quantify the proportion of hate and dangerous speech among the total conversations occurring on a social networking platform. These figures can offer a more solid grounding for shaping the debate on the possibilities and risks of social media in Ethiopia, but also create the foundations for comparative research that can quantify the prevalence and significance of these forms of speech in different national contexts and explore the reasons behind similarities and differences.


You may find the reports here. Mechachal – Online Debates and Elections in Ethiopia. Report One: A preliminary assessment of online debates in Ethiopia and Report Two: Discussing politics and history in social media


You are free to re-publish - just give credit to the author and to HornAffairs with a link.

Daniel Berhane
About Daniel Berhane (935 Articles)
Daniel Berhane