In collaboration with Matteo CM Casiraghi and  Zoé Christoff, I investigate when and how reaching agreement is possible in today’s polarised society. First, I identify argumentative and communicative strategies that contribute to issue polarisation, i.e., belief-based group formation, between members of opposing groups. Second, I examine whether low levels of polarisation support more solution-oriented communication than allowing polarisation to emerge through contrasting, yet potentially constructive, exchanges. I address these questions through interaction experiments involving real-life dyadic and triadic, spoken and written exchanges between polarised individuals, measuring both polarisation and dialogue outcomes. To examine how uncertainty expressed by different sources (AI, social media, news) interacts with political ideology to shape polarization, Casiraghi and I conduct survey-based research. In parallel, Christoff and I explore how these findings can inform computational agent-based models to simulate communicative strategies at the group level.

In collaboration with Dennis Ulmer, Christian Hardmeier and Ivan Titov, we argue in our position paper that large language models (LLMs) should communicate uncertainty in more human-like ways, through calibrated verbal cues like hedging and epistemic markers, to foster user trust, reduce overreliance, and improve transparency. Drawing on interdisciplinary research, we advocate for an “anthropomimetic” approach that aligns model outputs with human communicative norms and cognitive expectations

In collaboration with Bodo Winter and Stephanie Solt, we investigate manual co-speech gestures as a means to empirically probe aspects of Generalised Quantifier Theory, such as monotonicty and vagueness. Our research aims to deepen our understanding of how quantifiers such as several, some and many are cognitively conceptualised and interpreted in terms of inferred quantities. For more information, have a look at our OSF project.

The interpretation of quantifiers such as several and some can be distributive – the predicate is applied to each individual separately – or collective – the predicate is applied to a group of individuals as a whole. Together with Taknoubu Nakamura, we investigate how comprehenders disambiguate the interpretation of several and some by considering potential biases (e.g., to assign several a distributive reading) and ways in which grouping versus individuating co-speech gestures may facilitate the disambiguation process. For more information, see our OSF project.

In collaboration with Dimitris Serafis, Janina Wildfeuer and Kun He, we explore how right-wing political figures in the Netherlands leverage multimodal strategies to advance their perspectives and appeal to the electorate. We do so by utilizing a multimodal annotation framework to analyse the complex interplay of verbal and non-verbal elements in videos by different political candidates in the Netherlands; and employing multimodal argumentation theory to investigate how their use of such a complex interplay ends up triggering argumentative inferences that lead to the reinforcement of Dutch far right political rhetoric. 

Together with David Neequaye, we examine the mechanisms by which interviewees in interrogations mentally organize information when deciphering what an interviewer wants to know. The overarching idea is that such a process stems from the extent to which an interviewer’s question specifies an objective. See our OSF project for more information.

Together with Bodo Winter and Thordis Raja Wobith, we recently started looking into  how speakers exploit cognitive biases like “denominator neglect” – the tendency to ignore the overall population and focus on individual cases – to manipulate listeners. For example, during COVID-19, comparisons such as “Germany has 100.000 cases compared to the UK’s 92.000” ignored population sizes. This omission could reflect the speaker’s biases or a strategic attempt to mislead the public about which country fared better. 

Together with Greg Woodin and Bodo Winter, we investigate the use of numbers. We found that numbers are often used approximately in American and British English, in line with results from the literature on numerical cognition. Our corpus analyses show that ‘rounder’ numbers are used more frequently at higher magnitudes, revealing increasingly imprecise use of larger numbers. Some measurement scales have different ‘round’ numbers (e.g., multiples of 12 for hours), but these scale-specific trends mostly disappear at higher magnitudes. Our analysis of jigsaw puzzles finds direct evidence for increased rounding for larger numbers: as these puzzles grow larger, the stated number of pieces on the packaging (e.g., 13,200) differs more frequently, and by a greater extent, from the actual number (e.g., 13,224). Corpus analyses also demonstrate that broadeners such as ‘about’ and ‘approximately’ are used more frequently with round numbers, especially at relatively high magnitudes, reinforcing their vague usage.  These results highlight that number use depends on roundness, magnitude, and measurement scale. Whether a number is interpreted by default as precise or approximate may be contingent on these three factors. See our OSF project for more info.