American psychologists: the mere detection of hateful language is not a real solution

Scholars in the field of Natural Language Processing (NLP) have in recent years focused on improving the automated detection of hate speech so such language can be removed from online spaces. But some researchers have argued that mere detection is not a real solution – that

NLP can and should be used to investigate the roots of hateful language. Some people use language toward hateful ends, inciting violence and genocide, intimidating and denigrating others based on their identity. Despite efforts to better address the language of hate in the public sphere, the psychological processes involved in hateful language remain unclear. Moral values such as purity and loyalty have been linked by antisemites such as the Nazis, for example, with hateful language. Only recently have researchers begun to propose that violence and prejudice may have roots in moral intuitions.

Psychology Prof. Morteza Dehghani and colleagues at the University of Southern California conducted three studies of hateful language, looking at speeches and texts written by leaders of the Nazi party between 1933 and 1945; hateful slurs in large text corpora across 25 languages; and contemporary hate speech from 2018 on the far-right social-media platform Gab. In each analysis, the authors used NLP to situate hate speech in the context of moral motivations, worldviews, and rationales.

The researchers published their findings in the journal PNAS Nexus under the title “The (moral) language of hate.” 

“We aim to supplement existing qualitative research on the rhetorical aims of Nazi propaganda with a quantitative text analysis, which specifically emphasizes its possible moral component,” they wrote. “Qualitative research on the propaganda of Nazi Germany, which is argued to be essential in motivating those who implemented the mass murder of European Jews and other victims, has described the rhetoric split along moral lines of right or good (Germans) and wrong or evil (Jews), with language that dehumanized and instilled fear of the out-group Jews, while glorifying and sanctifying the in-group of Aryan German.”

They investigated moral language directed at in-group members (Germans) and out-group members (Jews). “We hypothesize that there is a relationship between hate and morality in language, with hateful language and moral language being concomitant. Based on prior research, we also predict that purity, loyalty, and authority will be associated with the Jewish out-group. Given the novelty of focusing on the in-group in studies of hateful language, we make no hypotheses about the morality of in-group rhetoric other than that we expect them to mirror the pattern found in out-group rhetoric,” they concluded.

The power of language

The power of language to incite hatred and spur violence is as clear today as it has been throughout history – propaganda in print and on the airwaves was used by Nazi leaders to turn a nation to genocide and hateful extremists in Rwanda spurred a genocide against the minority Tutsi population by dehumanizing and incendiary rhetoric on the radio. Clearly, language is too often subverted by hateful individuals and groups to harm outgroup members.

They gave another example. “Do not relent in purifying and cleansing the Arabian Peninsula of polytheists, heretics, and apostates,” said Osama bin Laden, the notorious founder of the Islamist terrorist organization al-Qaeda and mastermind of numerous terrorist attacks against the US and other Western powers, including the 2000 suicide bombings in New York and Washington.

The team used Moral Foundations Theory – which suggests moral sentiments can be classified into a small number of core values, including care, fairness, loyalty, authority, and purity – as their overarching framework. The authors found that Nazi propaganda appealed to the ideal of purity in the sense that members of out-groups were depicted as impure and polluting.

In the large multi-lingual corpora, hateful language was highly connected with words related to the ideal of loyalty. And on Gab, purity was again the most invoked value in hateful posts. Morality and hate may be deeply linked in the human psyche, according to the authors. Alternatively, or in addition, moral arguments may be used to legitimize hateful opinions. The authors called for additional research into the link between morality and hate.