Monitoring Online Hate Speech in the Kyrgyzstan Election
With contributions from the Westminster Foundation for Democracy’s team, Ben Graham Jones – a freelance consultant on elections and an expert in social media monitoring methodologies – sheds lights on the innovative approaches to monitor instances of hate speech online pioneered in the framework of the Media Dialogue project by Kloop media and the Independent Union of Journalists of Kyrgyzstan.
The Covid-19 pandemic has accelerated the digitisation of election campaigns. This has accelerated challenge facing election analysts to effectively monitor the online environment. Yet in parallel with ongoing digitisation, pioneers within the field of election observation are developing innovative new methodologies to observe these campaigns. We are fortunate to be able to build upon the strong foundations laid in recent years by organisations such as NDI/IFES, the EU and the Westminster Foundation for Democracy.
Yet whilst methodological reformers have long explored innovative approaches to examining phenomena such as online ads or digital media, one area that has received less attention within the elections field is the usage of innovative online tools to analyse hate speech on social networks. Perhaps as a result of the assumption that it is innately more subjective than dimensions such as online ad expenditure, comparatively few attempts have thus far been made to use innovative means to systematically track online hate speech within the domain of election observation.
Often, methodological inspiration can be found by looking at particular innovators operating within a single country, and then reflecting on how these methodologies might be applied in other contexts. It is in this context that it is a great pleasure to have been involved in supporting two innovative organisations in Kyrgyzstan, the Independent Union of Journalists and Kloop. As part of a consortium led by the European Partnership for Democracy with funding from ALDA – the European Association for Local Democracy, both implemented new approaches that challenge the boundaries of what has been achieved so far within the realm of election observation. In doing so, both combined innovative approaches with robust definitional frameworks inspired by the tried-and-tested approach of Article 19.
This has not been without challenges. Such is the nature of pioneering methodologies. For Kloop, one of the unique strengths of their approach has been how they have sought to deploy embeddings to observe Kyrgyz-language content. Yet the fact that Google Universal Sentence Encoder does not yet support the Kyrgyz language has meant that innovative alternatives had to be explored. For the Independent Union of Journalists, one challenge was how to present the results of an analytic methodology that has so far assessed more than 7000 articles. Yet a user-friendly interface, still a work in progress but already clear and well-defined, is making their work readily accessible.
In addressing these challenges, both organisations have made real progress that will have meaningful lessons for methodological reformers within election observation. The challenge, once the projects are finished, will be to connect their experiences with the wider debate on ways forward for both international election observation and domestic election observation in other countries. Indeed, finding ways to make the connection between CSO innovation and wider knowledge-sharing is a challenge that presents itself right across the field of democracy support.
The results of the project highlight the ongoing challenge hate speech poses in the democratic process. For example, impressively, IUJ’s methodology included the development of the first (to our knowledge) dictionary of terms associated with hate speech in the history of Kyrgyzstan. Developed in collaboration with an expert linguist, this helped the process of automation, thereby allowing the project to analyse a quantity of data that would have been unimaginable through manual means. It gave a sense of just how prevalent hate speech can be at election time.
In presenting such innovative new approaches, it is important to ensure analysts who will in time make use of the IUJ work can exclude legitimate political debate and critique from the dataset amassed through the methodology. To facilitate this, their work separated the incidences of hate speech by category groupings. IUJ analysts found that each of eight distinct types of hate speech were present in their sample, including incitement to violence. As other organisations look to their work, the challenge of ensuring that legitimate political discourse is not conflated with hate speech within automated analyses will be a major consideration, and emphasises the importance of pairing the power of automated data collection with human analytic interpretation who can verify and interpret the findings.
This election has also seen work undertaken not just to monitor, but to actively prevent hate speech. In parallel with the Media Dialogue project, project partner WFD collaborated with both the Central Electoral Commission of Kyrgyzstan and Civic Platform, a civil society organisation that supports the electoral process in Kyrgyzstan, to set defined standards against hate. This project saw the development of a memorandum of understanding and online consultations with political parties. The Memorandum outlines principles guiding the parties’ behaviour during the campaign ahead of the 4 October parliamentary election. A key element of the Memorandum was a commitment to avoid Hate Speech during the electoral campaign.
The Head of Central Election Commission of Kyrgyzstan, Nurzhan Shaildabekova Karmabekovna said: “… by signing the memorandum, political parties agree to multiply efforts to make elections fair and transparent. They also agreed not to use social media to denigrate opponents, and not to use Hate Speech”. It shows that important steps are being taken forward in both understanding the issue, but also in acting on it.
Whilst the projects are more pioneering steps forward rather than long-established finished products, significant progress has been undertaken. Whilst work is ongoing in the preparations for the rescheduled elections and in exploring how best to present the work undertaken so far, the steps forward in such a short space of time is impressive indeed. It is a reminder how in the ongoing debate on how best to monitor not just hate speech, but the whole array of online phenomena in elections, international NGOs can seek ideas and inspiration by looking at the work of civil society organisations. Keeping that conversation moving forward with fresh energy in both directions is key for continued progress on observation methodologies for the online dimension of elections.