Représentation des Connaissances dans un Langage Visuel Typé : une Première Approche

Florence Dupin de Saint-Cyr, Denis ParadeIC 2018 · OpenReview

TL;DR

This paper introduces VTL (Visual Typed Language), a novel visual language designed for representing knowledge. VTL aims to improve upon existing visual languages by adhering to principles of accessibility, navigability, and efficiency, using a combination of visual elements like shapes, colors, and icons to represent entities, actions, and conditions.

What Problem Does This Paper Address?

The paper addresses the challenge of effectively representing and communicating complex knowledge in a way that is intuitive and easily understood by humans. While text-based communication is prevalent, visual representations can often convey information more efficiently and memorably. Existing visual languages, such as concept maps or flowcharts, have limitations in terms of clarity, expressiveness, and adherence to cognitive principles of human perception. The authors identify a need for a visual language that is not only expressive but also adheres to principles derived from cognitive psychology and semiotics to ensure it is intuitive, efficient, and navigable.

Methods

The authors propose a set of formal principles and natural language postulates for an effective visual knowledge representation language. These principles include:

  • Accessibility: All information should be reachable.
  • Navigation: Users should be able to move between any pieces of information.
  • Efficiency of Access: Information should be quickly discoverable.
  • Entry Point: Each expression should have a single, unique entry point.

Based on these principles, they introduce VTL (Visual Typed Language). VTL categorizes elements into three types: entities, actions, and conditions. These elements can be generic or specific instances and are connected by typed relationships. The language leverages visual cues like shapes, colors, and icons, drawing inspiration from semiotics (e.g., Bertin's visual variables) and Gestalt psychology (e.g., proximity, continuity). They also propose a navigation mechanism that allows users to zoom in and out of representations, providing different levels of detail.

The paper also discusses the MOT (Modélisation par objets typés) language as a precursor, highlighting its strengths and weaknesses, and positions VTL as an improvement.

Datasets

Not discussed in the paper. The paper focuses on the design and principles of the VTL language itself, with examples created using the XMind software.

Key Insights and Ideas

  • Formalizing Visual Language Principles: The paper's core contribution is the formalization of principles for effective visual knowledge representation, moving beyond ad-hoc design to a more principled approach grounded in cognitive science.
  • VTL's Unified Approach: VTL integrates entities, actions, and conditions with typed relationships, offering a structured way to represent complex knowledge.
  • Leveraging Visual Semiotics and Gestalt: The design of VTL's symbols and their relationships is explicitly informed by principles of visual perception and semiotics, aiming for intuitive understanding.
  • Navigability as a Core Feature: The emphasis on zoomable and navigable representations addresses the issue of information overload in complex diagrams.
  • Bridging Visual and Formal Representations: The authors express a future goal of translating VTL into formal logic, enabling inference and verification.

Key Results

The paper presents the design of the VTL language and demonstrates its potential through examples. While no quantitative results are presented, the authors argue that VTL satisfies many of the proposed principles, including:

  • Accessibility and Entry Point: Ensured by the design of a single entry point and the ability to navigate to any element.
  • Navigation: Facilitated by the zoomable interface and typed relationships.
  • Efficiency of Access: Addressed by the hierarchical and navigable structure.
  • Visual Clarity: Achieved through distinct shapes for entities, actions, and conditions, and the use of icons and colors.
  • Double Coding: The combination of text and visuals supports the dual-coding theory, enhancing memorability.

Limitations and Future Work

  • Experimental Validation: The authors acknowledge that the proposed VTL language and its symbols have not yet undergone experimental validation with users. This is a crucial next step to confirm its intuitiveness and effectiveness.
  • Tool Development: While XMind was used for prototyping, the development of a dedicated graphical user interface (GUI) for VTL is planned.
  • Formal Translation: The paper outlines the intention to study the translation of VTL into non-visual formal languages to enable inference and coherence checks.
  • Completeness of Visual Variable Implementation: Some visual variables (like size) are noted as not being fully implemented yet.
  • Semic Relations: The paper mentions exploring more advanced semantic relations (like Allen's intervals or RCC8) and the potential for automatic translation of visual expressions.
  • Consistency in Contrasting Symbols: The authors note that the color choices for "entry" and "exit" doors are not consistently indicative of their opposing functions, a point for future refinement.

Summary updated

Abstract

fr Cet article présente un ensemble de principes qu'un langage visuel de représentation des connais-sances doit satisfaire. Nous avons introduit quatre principes formel (accessibilité, navigation, efficacité d'ac-cès, existence d'un point d'entrée) et sept postulats en langage naturel visant à contraindre la cohérence, compacité, navigabilité du langage par rapport aux attributs visuels (position, forme et couleur). Après une présentation du langage MOT, nous introduisons le langage VTL (Visual Typed Language) qui en est inspiré. VTL est un langage visuel typé qui satisfait la plupart des principes que nous proposons. Nous avons séparé les éléments du langage VTL en trois catégories : entités, actions et conditions. Les expressions de chaque catégorie sont soit des éléments génériques soit des instances. Ils sont reliés entre eux par des attributs dépendant de leur catégorie, ces liens permettent la navigation. Enfin, nous avons réalisé une maquette interactive à l'aide du logiciel XMind, initialement dédié aux cartes mentales.
BibTeX
@inproceedings{saintcyr2018representation,
  author = {Florence Dupin de Saint-Cyr and Denis Parade},
  title = {{Représentation des Connaissances dans un Langage Visuel Typé : une Première Approche}},
  booktitle = {IC},
  year = {2018},
  url = {https://openreview.net/forum?id=ArVsJrm64R},
}
Figure