Brief

Researchers have developed a new framework called inference-time argumentation (ITA) for claim verification, particularly useful when information is incomplete or conflicting. ITA is a trainable neurosymbolic system that uses formal argumentation semantics to guide Large Language Model (LLM) training. This approach allows models to generate arguments and assign them scores, which are then used to compute ternary predictions (true, false, or uncertain). The framework ensures that predictions are deterministically derived from explicit argumentative structures, offering more faithful explanations than post-hoc reasoning. AI

Researchers have developed a new neurosymbolic framework called Inference-Time Argumentation (ITA) for claim verification. This method trains large language models to generate arguments and assign them scores, which are then used to compute ternary predictions (true, false, or uncertain). ITA ensures that the final verdict is deterministically derived from explicit argumentative structures, offering more faithful explanations than traditional models. The framework has shown competitive performance against existing baselines on claim verification tasks. AI