Robust Symbolic Reasoning for Visual Narratives via Hierarchical and Semantically Normalized Knowledge Graphs

TL;DR

This paper presents a semantic normalization framework for hierarchical knowledge graphs in visual narratives, improving reasoning robustness and interpretability by consolidating semantically related actions and events.

Contribution

It introduces a novel normalization method for narrative graphs that reduces inconsistency and enhances reasoning in multimodal story understanding.

Findings

Normalization improves coherence in narrative reasoning tasks

Semantic clustering reduces annotation noise and redundancy

Framework maintains interpretability while enhancing robustness

Abstract

Understanding visual narratives such as comics requires structured representations that capture events, characters, and their relations across multiple levels of story organization. However, symbolic narrative graphs often suffer from inconsistency and redundancy, where similar actions or events are labeled differently across annotations or contexts. Such variance limits the effectiveness of reasoning and generalization. This paper introduces a semantic normalization framework for hierarchical narrative knowledge graphs. Building on cognitively grounded models of narrative comprehension, we propose methods that consolidate semantically related actions and events using lexical similarity and embedding-based clustering. The normalization process reduces annotation noise, aligns symbolic categories across narrative levels, and preserves interpretability. We demonstrate the framework on annotated manga stories from the Manga109 dataset, applying normalization to panel-, event-, and story-level graphs. Preliminary evaluations across narrative reasoning tasks, such as action retrieval, character grounding, and event summarization, show that semantic normalization improves coherence and robustness, while maintaining symbolic transparency. These findings suggest that normalization is a key step toward scalable, cognitively inspired graph models for multimodal narrative understanding.