NL4DV

NL4DV transforms natural-language queries into structured analytic specifications for data visualization, implemented in Python and producing JSON objects and Vega-Lite specifications.

Key Features:

• Python implementation: NL4DV is implemented in Python.
• Input handling: Accepts a tabular dataset alongside a natural language (NL) query as input.
• Output format: Generates a JSON object that encapsulates data attributes, analytic tasks, and Vega-Lite specifications.
• Natural language mapping: Maps natural-language expressions to analytic tasks and visualization encodings.
• Vega-Lite integration: Produces Vega-Lite chart specifications as part of its output.
• Multimodal support: Can incorporate speech input for multimodal visualization systems.
• Jupyter rendering: Produces specifications that can be rendered within Jupyter notebooks.

Scientific Applications:

• Visualization in Jupyter Notebooks: Render visualizations directly from natural language queries in Jupyter notebook environments.
• Development of NLIs for Vega-Lite charts: Enable creation and editing of Vega-Lite charts via natural language interfaces.
• Recreation of data ambiguity widgets: Support recreation of widgets similar to those in the DataTone system to address ambiguities in data interpretation.
• Multimodal visualization systems: Integrate speech input to enable multimodal interaction with visualizations.

Methodology:

NL4DV parses a natural language query together with a tabular dataset and emits a JSON object containing data attributes, analytic tasks, and Vega-Lite specifications; it can also accept speech input for multimodal systems.