[lamarr-nlp] Guest Talk by Liesbeth Allein from Ghent University | Uncovering Implicit Causal Thinking in Climate Discourse

Wednesday 14 Oct 2026, 10:00 → 11:00 Europe/Berlin

As part of the Lamarr NLP Colloquium, we have the pleasure to host Liesbeth Allein from Ghent University. Liesbeth will give a talk on implicit causal reasoning in climate discourse.

Title: Uncovering Implicit Causal Thinking in Climate Discourse

Abstract:

Causal reasoning lies at the heart of how people understand and debate complex societal issues like climate change. While prior work in NLP has primarily focused on identifying explicit cause-effect relations in text and building causal graphs from these observed relations, much of human causal reasoning relies on implicit, often contested, causal knowledge and the underlying causal mechanisms that connect causes to effects. In this talk, I present a multidisciplinary line of research at the intersection of linguistics and NLP that uncovered and modeled the implicit cognitive causal mechanisms that underlie causal discourse about climate change.

I will first briefly open the linguistic chapter. Here, we mapped explicit patterns of causal expression in climate-believer and climate-skeptic Reddit communities. Do these communities discuss causality differently? And how do they assign agency and responsibility? Our findings showed the importance of common ground, where shared causal assumptions shape argumentative dynamics. Both agreements and disagreements about these assumptions influence public debate, yet they are rarely articulated explicitly.

The NLP chapter, which forms the main part of the talk, seeks to recover the implicit causal thinking that underlies explicit causal statements (e.g., climate change causes flooding). To that end, we introduced implicit causal chain discovery: the reconstruction of intermediate causal mechanisms linking an explicitly stated cause (climate change) and effect (flooding) in text. These mechanisms, which often take the form of causal chains, represent the cognitive causal pathways that people build in their minds (climate change -> higher temperature -> more intense rainfall events -> flooding) to make sense of the causal relation presented to them (climate change -> flooding). Using climate discourse as a testbed, we first assessed the ability of LLMs to generate such causal chains. We examined whether these models perform genuine causal reasoning or rely primarily on memorized associative patterns and prompt-sensitive heuristics. We continued to design end-to-end and iterative approaches for constructing implicit causal graphs that capture multiple interacting causal pathways. Graph construction was additionally constrained to promote event diversity and avoid cycles in the graph, and we tested whether digging in the causal knowledge of multiple LLMs during construction improves graph quality. Finally, we designed a transparent database-based framework for evaluating the causal validity of the constructed causal graphs. Since no reliable resource existed, we built a database of scientifically vetted cause-effect pairs that we manually extracted from United Nations reports on climate change. The manual annotation effort produced a challenging dataset for causal discovery from science-for-policy text, with unique annotations for cause-effect correlation and spatiotemporal contextualization.

Keywords: causality, climate change, causal reasoning, causal chain discovery, causal graph construction, causal discovery from text

 

Bio:

Liesbeth Allein is a postdoctoral researcher at the Artificial Intelligence and Data Analytics research group at Ghent University (BE). She currently holds a fellowship from Research Foundation - Flanders (FWO), in which she advances language models with human social reasoning skills for socially-aware decision-making. Her main research interests lie in the broader field of natural language processing, with a primary focus on implicit language understanding. Previously, she held positions at KU Leuven (BE) (postdoctoral researcher; 2024-2025) and the Joint Research Centre of the European Commission (IT; doctoral grantholder, 2020-2021). She has also visited NLP groups at Ghent University (BE) (2024-2025) and University of Copenhagen (DK) (2020). She obtained her PhD in Engineering Science: Computer Science at KU Leuven (BE), under the supervision of Prof. Dr. Marie-Francine Moens (2024), titled Machine Learning Models for the Verification of Information and Detection of Disinformation. Liesbeth further holds a M.Sc. Artificial Intelligence (2019) from KU Leuven (BE), M.A. Multilingual Communication (2017) and B.A. Applied Linguistics (2016) from Ghent University (BE).

 

Looking forward to your participation.

Date: Wednesday, October 14, 2026
Time: 10:00 - 11:00 am (CEST).
Location:  Friedrich-Hirzebruch-Allee 6, 53115 Bonn, Germany
Room: 2.122 + Zoom
Zoom:  https://uni-bonn.zoom-x.de/j/63819604806?pwd=64PSGa9HyTym9j1bjy6jhcJF3eHebi.1