References & Citations
Computer Science > Computation and Language
Title: PInKS: Preconditioned Commonsense Inference with Minimal Supervision
(Submitted on 16 Jun 2022 (this version), latest version 13 Aug 2023 (v2))
Abstract: Reasoning with preconditions such as "glass can be used for drinking water unless the glass is shattered" remains an open problem for language models. The main challenge lies in the scarcity of preconditions data and the model's lack of support for such reasoning. We present PInKS, Preconditioned Commonsense Inference with WeaK Supervision, an improved model for reasoning with preconditions through minimum supervision. We show, both empirically and theoretically, that PInKS improves the results on benchmarks focused on reasoning with the preconditions of commonsense knowledge (up to 40% Macro-F1 scores). We further investigate PInKS through PAC-Bayesian informativeness analysis, precision measures, and ablation study.
Submission history
From: Muhao Chen [view email][v1] Thu, 16 Jun 2022 04:50:00 GMT (850kb,D)
[v2] Sun, 13 Aug 2023 20:53:43 GMT (861kb,D)
Link back to: arXiv, form interface, contact.