Research programme
CArLA connects argumentative causal discovery, graded reasoning, scalable graph learning and language-model knowledge with methods for contestability and redress.
Combines statistical evidence and expert knowledge through assumption-based argumentation, making causal conflicts explicit and returning supported alternative structures.
Contribution to CArLA: This is the methodological foundation of CArLA: causal structures are not presented as unexplained outputs, but as conclusions supported by inspectable and contestable arguments.
Assigns dialectical strength to assumptions, enabling more nuanced comparisons than binary acceptance alone.
Contribution to CArLA: Gradual evaluation gives CArLA a route beyond yes-or-no decisions, supporting explanations that expose relative support and uncertainty when evidence is noisy or incomplete.
Uses heterogeneous graph neural networks to approximate credulous acceptance and reconstruct stable extensions for large ABA frameworks.
Contribution to CArLA: The work addresses the scalability objective by providing learned approximations while retaining a route back to valid argumentative extensions.
Uses language models as imperfect experts that propose causal constraints, then evaluates those constraints alongside statistical evidence through Causal ABA.
Contribution to CArLA: This extends CArLA with a scalable source of external knowledge without treating language-model output as unquestionable truth: proposals remain attributable and contestable.
Recasts the release of inconsistent causal constraints as a cost-aware repair problem using optimal minimal correction sets.
Contribution to CArLA: Optimal correction sets make inconsistency handling principled and auditable: CArLA can explain which constraints were released, why they were selected, and which conclusions remain supported.
