Translating Visualization Queries into Natural Language
The utility of visualization tools is increasingly growing through providing highly interactive visual exploration and expressive querying on multidimensional data. However, as tools get more sophisticated, their usability considerably decreases. Therefore, there is a need to make these sophisticated visual and analytical tools more understandable. The goal of this project is to exploit natural language familiarity to directly facilitate individual and collaborative use of visual analysis tools.
To fulfill this goal, we will generate a query-to-question interface, which translates user interaction into natural language. This makes visualization states more understandable, and reveals how the tools themselves work. We will utilize natural language generation (NLG) techniques to automatically generate questions based on user interactions.
For now, we have constrained our research on coordinated multiple views, which are increasingly used to design tools for highly interactive visual exploration and analysis of multidimensional data.
At first, we focused on cross-filtering aspects and generated a domain-dependent interface that used designer predefined questions to reflect user's interactions. Now, we are generalizing the idea using NLG and defining more general structure for the questions and fianlly will generate questions automatically.