Expanding the Horizon of Linear Temporal Logic Inference for Explainability

Linear Temporal Logic (LTL), a logical formalism originally developed for the verification of reactive systems, has emerged as a popular model for explaining the behavior of complex systems. The popularity of LTL as explanations can mainly be attributed to its similarity to natural language and its ease of use owing to its simple syntax and semantics. To aid the explanations using LTL, a task commonly known as inference of Linear Temporal Logic formulas, or LTL inference in short, has been of growing interest in recent years. Roughly, this task asks to infer succinct LTL formulas that describe a system based on its recorded observations. Inferring LTL formulas from a given set of positive and negative examples is a well-studied setting, with a number of competing approaches to tackle it. However, for the widespread applicability of LTL as explanations, we argue that one still needs to consider a number of different settings. In this vision paper, we, thus, discuss different problem settings of LTL inference and highlight how one can expand the horizon of LTL inference by investigating these settings.