XISM: an eXploratory and Interactive Graph Tool to Visualize and Evaluate Semantic Map Models

2025-07-05

AI Summary

Semantic map modeling in linguistic typology faces bottlenecks including low efficiency, insufficient visualization, and lack of standardized evaluation tools. This paper proposes a human-in-the-loop framework for semantic map construction, integrating data-driven, graph-theoretic top-down generation algorithms with expert linguistic knowledge to automatically derive candidate semantic maps from user-provided data. The system features an editable, interactive interface supporting dynamic refinement and real-time visualization. It incorporates multidimensional evaluation metrics—including coverage, parsimony, and cross-linguistic consistency—to enable rigorous, interpretable assessment. Empirical results demonstrate substantial improvements in modeling efficiency and transparency. The open-source implementation constitutes the first semantic map modeling platform that simultaneously ensures automation, verifiability, and expert intervention, thereby advancing both typological and computational linguistic research.

Technology Category

• Natural Language Processing: Lexical Semantics and Morphology
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Knowledge Representation and Reasoning: Knowledge Representation Languages

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

Semantic map models represent meanings or functions as nodes in a graph constrained by the local connectivity hypothesis, with edges indicating their associations. Widely used in typological linguistics, these models compare interrelated meanings across languages. Traditionally built manually in a bottom-up manner, they are inefficient for large datasets and lack visualization and evaluation tools. This paper introduces XISM, an interactive tool based on our prior algorithm, which constructs semantic maps from user data via a top-down approach, displays candidate maps, and evaluates them using multiple metrics. Users can refine maps by editing edges, combining data-driven efficiency with expert knowledge. This human-in-the-loop design benefits both typologists and computational linguists. The system https://770103knev48.vicp.fun/ and a demonstration video https://youtu.be/S-wsVDF2HSI?si=1OrcF41tRznaifhZ are publicly available.

Problem

• Lack of efficient tools for visualizing semantic map models
• Manual construction of semantic maps is time-consuming for large datasets
• Absence of interactive evaluation metrics for semantic map models

Innovation

• Top-down algorithm constructs semantic maps
• Interactive tool visualizes candidate maps
• Human-in-the-loop refines maps with metrics