Type-Directed Synthesis of Visualizations from Natural Language Queries

Explore a groundbreaking technique for automatically generating visualizations from natural language queries in this 16-minute conference talk from ACM SIGPLAN's OOPSLA. Discover how the proposed method utilizes program synthesis to parse natural language queries into refinement type specifications using the intents-and-slots paradigm. Learn about the type-directed synthesis approach that generates visualization programs likely to meet user intent. Understand how the refinement type system captures useful hints from queries and rejects visualizations violating design guidelines. Gain insights into the implementation of these ideas in the Graphy tool and its evaluation on the NLVCorpus dataset. Examine the experimental results showing Graphy's superior performance compared to state-of-the-art natural language-based visualization tools, including transformer and rule-based approaches.