Opinion or hate speech?

In Belgium, discriminatory comments are punishable by law. But what is to be done when the words are at the borderline between opinion and hate speech? The team of UCLouvain linguist Barbara De Cock conducted a study with VUB on the linguistic characteristics of the words politicians used on social networks before and during Belgium’s May 2019 election campaign. Strategies that incited hatred and escaped justice were clearly identified.

In recent years, tweets and Facebook posts have become headlines, even for people who don’t use social networks. Political figures and anyone else express their thoughts and opinions freely, and sometimes go off the rails. Some use hate speech, i.e. messages that publicly encourage people to discriminate or commit acts of hate and violence against individuals on the basis of ‘protected criteria’ (sex, nationality, race, etc.). This is prohibited and punishable by law. Others use social media to express an opinion, without a sweeping value judgment of a targeted group. In between hate speech and opinion is a ‘grey area’, which researchers at UCLouvain and VUB studied following a call for proposals by Unia, a Belgian institution that combats online hate speech.

What do we analyse?

Have you ever read a social network post that fuelled or justified discrimination or hatred towards certain groups without explicitly encouraging it? There are many such posts. What about in the political realm? For every Belgian party with seats in Parliament during the legislature preceding the May 2019 election, Dr De Cock and her team collected the tweets and Facebook posts of four accounts: those of the party president, the party generally, and two prominent party members linked to topics Unia addresses, such as discrimination. Analysing both French and Dutch posts, the team concentrated on two periods: a non-campaigning period from 26 January to 26 February; and a campaigning period from 26 April to 26 May.

Collecting and filtering data

Most of this research work was data collection. ‘Our research focused on speech concerning groups or individuals characterised by their religious or philosophical conviction, ethnic origin or sexual orientation, that is, the protected criteria on which Unia bases its mission’, Dr De Cock explains. First, the researchers used the Facepager application to collect Facebook data and CENTAL to download tweets. Secondly, they filtered ‘grey area’ speech. ‘We had to select speech that implicitly or explicitly created a demeaning or even threatening representation of a given social group, potentially justifying discrimination against or hatred of it,’ Dr De Cock says. To do this, the team’s French and Dutch-speakers scanned all the messages, then selected those that were relevant, by asking of each: Is this an opinion? Hate speech? Or somewhere in between, which can be perceived by some as discrimination? ‘This selection was manual because the linguistic strategies used are often implicit, so it’s difficult to automate.’

Strategies for evading punishment

These different stages of work led our researchers to several observations and conclusions. The first general observation: the messages contained very few strategies that incited hatred openly, and are based mainly on implicit and indirect language. ‘They suggest more than they claim that certain groups represent a problem or even a danger,’ Dr De Cock explains. These messages use several linguistic strategies.

• Create a clear opposition between the poster’s group (endogroup) and another group by suggesting homogeneous composition of each. For example, Etienne Dujardin (MR) retweeted a sentence by French politician Henri Guaino: ‘The West’s great weakness is not in number or size but self-denial.’ The quotation assumes that ‘the West’, and therefore ‘Westerners’, have their own, intrinsic, permanent identity, which they tend to deny.
• Represent the exogroup in a negative way on the basis of assumed characteristics by associating it with negative phenomena or acts. Yasmine Dehaene (PP) posted on Facebook: ‘Dealing drugs on every street corner ... residential buildings without a single European name.’ The juxtaposition of this information suggests a link between drug-dealing and being non-European.
• Disqualify opposing discourses. Posted on the official Facebook account of the Listes Destexhe party: ‘The left will always surprise us by demonstrating an unfailing creativity for justifying these figures ... in order to seduce a voting bloc to whom it owes so much.’
• Use hyperbole. Referring to Brussels’s streets, Dehaene posted on Facebook: ‘I can tell you that it’s catastrophic.’
• Remain vague by using ‘we’ and ‘you’ forms. In the same Facebook post, Dehaene wrote: ‘neighbourhoods where they hunt us’. Who are ‘they’ and ‘us’? The choice of the verb is precise, given ‘hunting’ evokes an intentional aggressive act.

No grey area messages during the campaign?

The initial research question posed by Unia was: Were there more grey area messages before or during the election campaign? Dr De Cock replies, ‘In general, politicians posted more frequently during the campaign but no longer posted hate messages. This could be explained in part by the constant campaign mode since the elections of October 2018, owing to the government’s fall.’ Were there more grey zone messages in the country’s north or south? ‘In our collection, there were more in Flanders because there were more tweets in general for the selected period.’ On the French-speaking side, such messages were produced mainly by Listes Destexhe and Parti Populaire candidates but also appeared in the discourse of politicians in other French-speaking parties. On the Dutch-speaking side, grey area messages came overwhelmingly from one party, Vlaams Belang, with the N-VA a distant second. What did Dr De Cock learn from this experience? ‘There were relatively few grey area messages in the discourse of prominent political figures. It seems they’re aware of the legal limits.’ Now that their research report has been published, the team is already thinking about the future. ‘We’re collaborating with Unia to integrate these results into their work, for example in training, collaborations, and processing reports of discriminatory messages.’