“…However, one of the greatest vulnerabilities acknowledged in these automatic detection systems is the lack of biunivocal correspondence between expression and meaning (Pisarevskaya, 2017;Rubin and Lukoianova, 2015), and more importantly the failure to take into account other dimensions and types of manipulation, analyzed in the rhetorical and dialectical tradition under the label of "fallacies" (even the ones that are essentially related to hate speech, see Habernal et al, 2018). In this picture, the traditional fallacies become extremely relevant for detecting the hidden side of manipulative discourse (Walton, 1987(Walton, , 1999, namely the information taken for granted by the speakers and not shared by the audience (Lombardi Vallauri et al, 2020;Lombardi Vallauri and Masia, 2014;Macagno, 2018b).…”