Data visualization serves as a crucial tool for communicating important information in our society. Yet, as visualizations grow more complex, they become less accessible to individuals with visual impairments. Traditional accessibility approaches like alternative text and data tables often fall short of capturing the full potential of data visualization. To bridge this gap, we introduce VizAbility, a novel multimodal accessible system that combines keyboard navigation with conventional interaction, enabling individuals with visual impairments to actively engage with and explore data visualizations. VizAbility utilizes an LLM-based pipeline, seamlessly integrating data, chart structures, user locality, and web-based information to provide comprehensive answers. Our quantitative evaluation validates the LLM-based question-and-answer pipeline, and a user study involving six participants underscores the promising potential of VizAbility's multimodal approach. We discuss how current visualization tools can integrate VizAbility to enhance the accessibility of data visualizations online.

翻译：摘要：数据可视化在社会中扮演着传递重要信息的关键角色。然而，随着可视化日益复杂，视障人士对其的可及性逐渐降低。传统的无障碍方法（如替代文本和数据表格）往往难以充分体现数据可视化的全部潜力。为弥补这一差距，我们提出VizAbility——一种结合键盘导航与常规交互的新型多模态无障碍系统，使视障人士能够主动参与并探索数据可视化。该系统采用基于大语言模型（LLM）的流水线，无缝整合数据、图表结构、用户位置及网络信息，以提供全面的回答。定量评估验证了基于LLM的问答流水线的有效性，而涉及六名参与者的用户研究则凸显了VizAbility多模态方法的巨大潜力。本文进一步探讨了当前可视化工具如何集成VizAbility，以提升在线数据可视化的无障碍性。