*

Umati : Monitoring Online Dangerous Speech

October 2012 Findings

Umati: Monitoring Dangerous Speech Online Introduction

Hate speech has garnered growing interest in Kenya, especially since the 2007 Post Election Violence, owing to its realised potential to stir or promote violence against targeted groups of people. The current denition of hate speech, according to the National Cohesion and Integration Commission Act of 2008 is speech that which advocates or encourages violent acts against a specic group, and creates a climate of hate or prejudice, which may, in turn, foster the commission of hate crimes. Due to this vague denition, there has been an escalating demand from peace-building organisations, politicians, government officials and the general public on how to dene, identify, mitigate, report and deal with hate speech. This need for less ambiguity in the denition of hate speech motivated the project to rely on a more actionable denition, and hence the monitoring of dangerous speech. In a nutshell, dangerous speech is hate speech with a potential to cause violence. Professor Susan Benesch of American University (Washington, DC, USA), denes dangerous speech as a subset of hate speech, and comprising of ve criteria:
1. A powerful speaker with inuence over an audience; 2. An audience vulnerable to incitement; 3. Content of the speech that may be taken as inammatory; 4. A conducive social and historical context of the speech; and 5. An inuential means of disseminating the speech.

Goals of Umati Following the need to dene, identify and deal with hate speech, the goals of the Umati project are: To set a denition of hate/dangerous speech that can be incorporated into the constitution. To forward incidences of dangerous speech to Uchaguzi to limit further harm. To dene a process for election monitoring that can be replicated elsewhere. To further civic education on hate speech.

Umati: Monitoring Dangerous Speech Online

Why Online? While most projects related to hate speech have been looking at mainstream media, we are aware of the inuencepositive and negativethat New Media such as the blogosphere and online forums had on the 2007 Post Election Violence in Kenya. Therefore, our agship project seeks to monitor and report, for the rst time, the role New Media plays on a Kenyan election. Our project will have citizens at the core and use relevant technology to collect, organise, analyse, and disseminate the information that we receive. Monitoring Process Over a period of 10 months beginning September 2012 and ending June 2013, the Umati project will monitor online content and record incidences of hate and dangerous speech. This process is being carried out by ve monitors, representing the ve largest ethnic groups in Kenya to enable the translation of cited incidences from vernacular to the countrys official language, English. Monitors use an online categorisation process that enables them to sort each collected statement into its respective category. Impact
Category One: Offensive speech Category Two: Moderately Dangerous speech Category Three: Extremely Dangerous speech

The three hate speech cateogories, in order of severity, are:

We hope that the work of this project will lead to the inclusion of a

more elaborate denition of illegal speech in the current constitution of Kenya, and that ndings will be used to educate the Kenyan public on what type of speech has the potential to disrupt peace and security in the country. Through this project, we aim to create a process that can be replicated in other countries to monitor dangerous speech leading up to pivotal national events, such as elections and referenda. Outputs The following section presents a consolidation of all incidences of hate speech that have been identied in Kenyas webosphere from three main sources: Social media (Facebook and Twitter), online blogs and comments sections of online newspaper

OCTOBER 2012 FINDINGS Moderately dangerous speech is the most rampant category of dangerous speech
Moderately Dangerous speech 35%

For the Umati project, the rst three criteria of dangerous speech were used to sort collected statements into the three hate speech categories. We were able to sort the statements by considering three questions about each of them:

n=774

- The observable response from the audience, it is little, mid or large. - How inammatory the statement is, whether mildly, moderately or highly inammatory. - The inuence the speaker has on the crowd, whether little, some or a lot of inuence.

Offensive speech 31%

Extremely The graph here shows that in October 2012, most statements fell in Dangerous speech the moderately dangerous speech category. These were statements 34% that called the audience to at least one call to action ( see page ) Important to note here is that extremely dangerous statements have the highest potential to stir violence. Examples of statements in October that fall in the three categories, are listed below:
Offensive speech: I saw it coming, its only in [political party]** that ppl are used and thrown away like condoms. Moderately dangerous speech: F*ck u all. [Religion1]** WIL Prevail. VICTORY Extremely dangerous speech: Hawa [tribe]** wana shida ya ulimi .watawekwa kamba wote.keep up [political party].

** In an effort to avoid fuelling hateful speech, we have deliberately omitted the naming of any tribes, political parties or politicians when writing this report. For example, when we quote statements verbatim from our study, we replace the named tribes with the terms [tribe1], [tribe2] etc.

OCTOBER 2012 FINDINGS Identiable commenters are most active users of dangerous speech
Surprisingly, the highest use of dangerous speech from the Kenyan online Umati is monitoring, is by identiable commenters. Identiable commenters are online users who leave comments in response to a Facebook post, an online news article, a forum or blog post. They are identiable in that they use their own name or a pseudoname. The lack of caution when speaking online suggests that the speakers are not considering the negative impact their statements could have, nor are they worried about being associated with the dangerous statements they make.

a blogger
6%

an identifiable commenter
53%

a politician
2%

an anonymous commenter
39%

n=776

an elder/community leader
0.05%

OCTOBER 2012 FINDINGS Most noted call to action is the call to discriminate
According to research carried out by Professor Susan Benesch of the American University, speech that contains certain calls to action can be deemed dangerous. Based on the audience and the speaker, these calls to action can generate varying degrees of incitement to violence. The calls are to discriminate, to steal, to riot, to beat, to forcefully evict and nally to kill. In the month of October, the most frequent call to action on monitored Kenyan blogs, newspapers, Facebook pages and tweets was to discriminate members of another group.
Discriminate
87%

Topic
Tribe Political party Religion Mombasa Republican Council (MRC) supporters

Frequency 417 166 10 4

Percentage 69 27 1.6 0.6

(out of a total of 609 incidences)

Other topics along which dangerous speech occured were against women, MRC supporters, homosexuals, MPs and politically backed criminal gangs. Some examples are listed below:

Forcefully evict
1%

Beat
1%

Riot
1%

Against people of a certain tribe: Tell those sleeping dogs to wake up to the reality, time is for [presidential candidate] to lead this country. So [tribe1], [tribe2] and all the Haters of [presidential candidate], no one from your tribe can lead this nation as per now so accept the fact and stop non-sensing here and if you want to see then wait for 2013. Against homosexuals: ..Gay?u shld b dead yesterday from a thorough beating...u dsav the worst u big Ass. I hate gay wil all my breath...may u b knocked by a lorry as u cross the road...proud to b a gay my vomit Against women: Hii mambo ya direct nomination for women really disheartens me,kazi ni kupeana kuma ndo ukue governor ama women rep, surely. when will men take us seriously? Of all women mps only a few who can even stand up and be counted...the rest are panty droppers.

Kill
10%

We also found that dangerous speech was centerd around particular topics with the top three being tribe, political parties and religion:

OCTOBER 2012 FINDINGS Events had a inuence on the occurence of dangerous speech online

In relation to who was being targeted, key events took place in October that increased the frequency of dangerous speech circulated in the online space. Notable events are: Feuds by rival gangs in Kisumu ( American Marines and China Squad) Re-emergence of Mungiki sect Ko Annans visit to Kenya Party hopping and political party alliances Ruto, Uhuru, Raila, Imanyara Hefty package to MPs Mombasa Republican Council ( MRC ) insecurity Sexual orientation of political candidate David Kuria 2013 Political elections Killing of Kenyan businesswoman in Southern Sudan Defection from ODM by Gitobu Imanyara Killing of Shem Kwega

OCTOBER 2012 FINDINGS Kill Most serious call to action highest with identiable speakers
Forcefully evict
Kill

Beat
Forcefully evict

Riot

Beat

*
Kill 6.5 0 0 4.3 0 89.1 0 0 0 0 100
6.5 0 0 4.3 0 89.1

Riot

Discriminate
Discriminate

a blogger a journalist
a blogger a journalist

Discriminate Discriminate 6.7 1

6.7 1

Riot Riot 0
0 0 0 0 0

0 0 0 0

Beat 0 0 0 0 0 100

Beat Forcefully evict 0 Forcefully evict 0 Kill 0 0 0 0 100

0 0 0 0 0 100

a politician a politician

1.71.7
38.6 38.6

an anonymous commenter an anonymous commenter an elder/community an elder/community leaderleader

an identifiable commenter

an identifiable commenter

0.70.7 51.3

51.3

100

100

The calls to action rank from the most severe which is to kill, to the least severe which is to discriminate. Commenters (both anonymous and identiable) reported the highest number of discriminatory statements. Identiable commenters showed a clear lead with statements that exhibited a call to kill another group. This was an interesting observation because the most serious call to action was highest amongst speakers that could be identied either by their Facebook names, or pseudonames used on public blogs and forums.

OCTOBER 2012 FINDINGS Calls to discriminate across the three categories of dangerous speech
Offensive speech Discriminate Riot Beat Forcefully evict Kill 98.80% 0.00% 0.00% 0.00% 1.20% Moderately Dangerous Speech 0.80% 1.10% 0.50% 0.00% 8.40% Extremely Dangerous Speech 75.20% 1.80% 1.80% 1.80% 13.30% Offensive speech

Given the context, speaker, and audience of the statement, the noted calls to action fell into different categories of dangerous speech. As a comparison, these three statements below are all calls to discriminate a particular tribe yet they fall in the three different categories of dangerous speech.

Offensive speech

Category 1: Offensive speech A [tribe1] will always be smarter and a good thinker that a [tribe2]... watch the outcome, [politician1] will run back like a hunted antelope after he has been duped by the [politician2]. Category 2: Moderately Dangerous Speech [tribe1] GO BURY YOUR MOTHERS ALIVE OR KIDNAP YOUR NIECES. SHAME. Category 3: Extremely Dangerous Speech Hawa [tribe1] wana shida ya ulimi .watawekwa kamba wote
Moderately Dangerous Speech 0.80% 1.10% 0.50% 0.00% 8.40%

Moderately Dangerous Speech

Extremely Dangerous Speech

Discriminate Riot Beat Forcefully evict Kill

Extremely Dangerous Speech 75.20% 1.80% 1.80% 1.80% 13.30%

This reiterates that the extremity of an inammatory comment, whether it falls under category 1 or 3, relies on a combination of facOffensive speech tors; 98.80% the inuence the speaker has over the audience, 0.00% how inciteful the statement is to the audience 0.00% and how harmful it is to the targeted group.
0.00% 1.20%

OCTOBER 2012 FINDINGS Number of identiable commenters increased with the severity of dangerous speech
From the above graph, two trends can be noted:

Anonymous commenters and identiable commenters were the most active amongst the speakers.

Offensive speech 39.90% 0.00% 57.20% 0.00% 0.40% 0.40% 2.10%

Moderately Dangerous Speech 55.90% 0.70%

an identifiable commenter 35.20% an elder/community leader

Extremely Dangerous Speech 59.40%

Offensive speech 0.80% 39.90%26.10% 57.20% 0.40% 0.00% 3.40% 0.40% 1.90% 0.40% 8.00% 2.10% 0.00%

Anonymous commenters reduced as the severity of hateful speech increased, while identiable commenters increased with the severity of dangerous speech. Identiable commenters were the largest users of extremely dangerous speech. This could be because an identiable commenter likely has greater inuence over an audience than an unidentiable commenter (the audience feels a stronger connection with an identied person rather than an anonymous stranger). The inuence of the speaker has an impact on our categorisation of the level of danger of the text.
Extremely Dangerous Speech 59.40% 0.80% 26.10% 0.40% 3.40% 1.90% 8.00%

10

dentifiable commenter

elder/community leader

Moderately Dangerous Speech 55.90% 0.70% 35.20% 0.00% 1.90% 0.00% 6.30%

anonymous commenter

ublic figure (includes media personalities)

0.00% an anonymous commenter

a public figure (includes media 1.90% personalities) a politician a journalist a blogger

olitician

urnalist

0.00% 6.30%

ogger

OCTOBER 2012 FINDINGS Most dangerous speech incidents were found in blog/forum comments

The greatest number of dangerous speech incidents was comments in response to a public blog article or forum. Facebook posts on a public group discussion or page closely followed.

Platform
A Facebook post in a private group/ page A Facebook post in a public group/page A blog article in a private blog/forum A blog article in a public blog/forum

Category
Offensive speech 1% 29% 0.4% 0.4% Moderately Dangerous Speech 16% 30% 0.4% 5% Extremely Dangerous Speech 22% 30% 3% 9%

11

A comment in response to a private blog article/forum A comment in response to a public blog article/forum A comment in response to an online news article A comment in response to an online news article or blog A tweet An online news article

3% 63% 2% 1% 0.0% 0.4%

6% 24% 10% 8% 0.0% 1%

5% 17% 2% 12% 1% 1%

OCTOBER 2012 FINDINGS Hallmarks of Dangerous Speech

From studying many examples from a variety of countries and historical periods, our project partner Prof. Benesch has identied tell-tale signs, or hallmarks, of Dangerous Speech. We adapted these hallmarks to the Kenyan context and searched for them within the statement we collected. The three hallmarks we used are : Compare a group of people with animals, insects or a derogatory term in mother tongue Suggest that the audience faces a serious threat or violence from another group, Suggest that some people from another group are spoiling the purity or integrity of the speakers group (e.g. characterize members of the other group as weeds spots, stains, rotten apples who may spoil a barrel, etc.)

Suggest that some people are spoiling the purity or integrity of another group

35%

12

1.0% 4.3%

26%
Suggest that the audience faces a serious threat or violence from another group

0.3%

0.3%

The diagram here shows that from the 302 sampled statements in October, most ( 35.4%), suggested that some people were spoiling the purity and integrity of another group.
* For further information and articles on the hallmarks and on Dangerous Speech generally, see www.voicesthatpoison.org

33%

Compare a group of people with animals, insects or a derogatory term in mother tongue

OCTOBER 2012 FINDINGS Hallmarks of Dangerous Speech across different speakers

Chart Title
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
1.0% 7.1% 2.0%

7.1%

2.0%

1.0%

Compare a group of people with animals, insects or a sects or a derogatory term in mother tongue 41.8% derogatory term in mother Compare a group of people with animals, insects or a derogatory term in tongue
mother tongue

48.0%
41.8%

Compare a group of people with animals, insects or a derogatory term in mother tongue

a blogger 48.0% a journalist a politician

11.4%

Suggest that the audience faces a serious threat or violence from another group

1.3%

Suggest that the audience faces a serious threat orSuggest that the audience faces a serious threat or violence from erious threat or violence violence from fromanother another group 25.3% group another group
11.4%

13

1.3%

62.0%
25.3%

a public figure (includes media personalities)

62.0% an anonymous commenter

an elder/community leader an identifiable commenter

a blogger

a journalist

2.8%

Suggest that some people are group spoiling theof purity or integrity Suggest that some people are spoiling the purity or integrity of another group iling the purity or integrity another group 37.4% of another group
4.7% 1.9%

Suggest that some people are spoiling the purity or integrity of another

4.7%

1.9%

a politician

a public figure (includes media personalities)

2.8%

37.4%

53.3%

53.3%

an anonymous commenter

an elder/community leader

an identifiable commenter

politician

a public figure (includes media personalities)

Across the speakers, the three hallmarks were contained in speech that we collected in October.

an anonymous commenter

an elder/community leader

an identifiable commenter

What is interesting to note here is that politicians were only present in the third hallmark i.e. suggest that some people are spoiling the purity or integrity of another group. while public gures were found to have said statements that suggested that the audience faces a serious threat or violence from another group.

Umati: Monitoring Dangerous Speech Online Notes

DEFINITIONS Discrimination: Discrimination is understood as any distinction, exclusion or restriction made on the basis of race. colour, descent, national or ethnic origin, nationality, gender, sexual orientation, language, religion, political or other opinion, age, economic position, property, marital status, disability, or any other status, that has the effect or purpose of impairing or nullifying the recognition or exercise, on an equal footing, of all human rights and fundamental freedoms in the political, economic, social, cultural, civil or any other eld of public life. Source: La Rue, F., (2012, September 7). Report of the Special Rapporteur on the promotion and protection of the right to freedom of opinion and expression, UN Doc A/67/357, p12. Dangerous speech: This is a term coined by Prof. Susan Benesch to describe incitement to collective violence that has a reasonable chance of succeeding, in other words speech that may help to catalyse violence, due to its content and also the context in which it is made or disseminated. This possibility can be gauged by studying ve criteria that may contribute to the dangerousness of speech in context: the speaker (and his/her degree of inuence over the audience most likely to react, the audience (and its susceptibility to inammatory speech), the speech act itself, the historical and social context, and the means of dissemination (which may give greater inuence or force to the speech). SOURCES Pictures on page 7 sourced from:
yara-rejoins.html. Accessed 12th December 2012.

Identiable Commenter: A person who responds to an online article, blog post or Facebook post who can be identied by a name, regardless of whether the name is real or fake.

14

http://www.thekenyanpost.com/2012/12/big-boost-for-raila-as-iman-

http://adrempress.com/ko-annan-claims-success-in-syria-peace-plan/. Accessed 12th December 2012. http://www.capitalfm.co.ke/news/2012/09/kisumu-falls-to-china-andamerica-gangs/2/?wpmp_switcher=mobile. Accessed 12th December 2012. http://www.the-star.co.ke/sites/default/les/styles/node_article/public/ images/articles/2012/01/12/34409/pwanisikenyawr.png#.UMrUyYP64rU. Accessed 12th December 2012.

OCTOBER 2012 FINDINGS

For more information on this project, contact

Umati Project Team iHub Research Nairobi, Kenya [email protected] www.research.ihub.co.ke | Twitter: @iHubResearch