Neurosymbolic Learning for Inference-Time Argumentation
The article discusses a new framework called inference-time argumentation (ITA) for claim verification in uncertain contexts. ITA utilizes a neurosymbolic approach to generate and score arguments, allowing for ternary predictions of true, false, or uncertain. The framework aims to provide transparent and faithful explanations for its predictions, improving upon existing models in terms of performance and interpretability.
- ▪Claim verification is critical in high-stakes areas like health and finance.
- ▪ITA is designed to handle incomplete or conflicting information by providing uncertain answers.
- ▪The framework optimizes argument generation and scoring during training to enhance prediction quality.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.20098 (cs) [Submitted on 19 May 2026] Title:Neurosymbolic Learning for Inference-Time Argumentation Authors:Gabriel Freedman, Adam Dejl, Adam Gould, Mansi, Lihu Chen, Jianqi Jiang, Francesca Toni View a PDF of the paper titled Neurosymbolic Learning for Inference-Time Argumentation, by Gabriel Freedman and 6 other authors View PDF HTML (experimental) Abstract:Claim verification is an important problem in high-stakes settings, including health and finance. When information underpinning claims is incomplete or conflicting, uncertain answers may be more appropriate than binary true or false classifications. In all cases, faithful explanations of the considerations determining the final verdict are crucial.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
