Paper: From Events to Reasoning — Hierarchical Rule Composition in the Event Reasoning System (ERS)

Forum topic · 小凯 · 2026-10-19

Summary

The Event Reasoning System (ERS) is a reasoning architecture by Cheng Lyu and Xueqi Wang (arXiv:2610.22403) in which the reasoning state is an explicit event graph rather than an opaque chain of conclusions. Semantic features nominate candidate objects and rule bindings, while declared rule and task conditions determine which operations are admissible, preventing candidates that violate period, unit, or evidence constraints from entering an accepted reasoning path. Low-level rules provide reusable local transformations, and recursively composable high-level control rules organize subgoals under the current goal, information gaps, and budget, so a finite rule library is rebound into paths never stored as complete solutions. The paper formalizes admissible transitions, task-dependent control, compositional expressivity, and dependency-preserving revision. Empirically, ERS builds dependency chains up to twelve steps, cuts irrelevant executed operations from 32 to 0 as lookahead grows to 16, produces no unsupported outputs across 24 negative cases, passes 12/12 frozen-rule checks and 6/6 fact-change cases, and uses 51.6% less inference wall-clock time than single-level global control on fourteen development tasks.

Paper Overview

Field: Machine Learning Authors: Cheng Lyu, Xueqi Wang Published: 2026-10-19 arXiv: 2610.22403

Abstract (Translated)

What a reasoning system emits is a conclusion; what it owes is a justification. Producing one requires every operation to be admissible on the facts actually available, the dependencies of intermediate results to survive to the end of the path, and the path to be revised when a premise changes. We present the Event Reasoning System (ERS), in which the reasoning state is an explicit event graph. Semantic features nominate candidate objects and rule bindings, and declared rule and task conditions decide which may execute, so candidates that violate the represented period, unit, or evidence constraints cannot enter an accepted reasoning path.

Low-level rules provide reusable local transformations, and recursively composable high-level control rules organize subgoals and legal paths under the current goal, information gaps, and budget; a finite library is thereby rebound into paths never stored as complete solutions. We formalize admissible transitions, task-dependent control, compositional expressivity, dependency-preserving revision, and the gap between reachable solutions and those a budgeted controller finds.

Key Results

On structured-event graphs, ERS:

Builds dependency chains of up to twelve steps
Cuts irrelevant executed domain operations from 32 to 0 as lookahead grows to 16
Produces no unsupported target output across 24 negative cases
With rules frozen, passes 12/12 focused checks (nine accepted derivations involving unseen input/path combinations, three justified refusals) and 6/6 fact-change cases
Uses 51.6% less inference wall-clock time (dynamic subgoal configuration) than single-level global control on fourteen jointly solved development tasks

--- *Auto-collected on 2026-10-19*