Inferencing underspecified natural language utterances in visual analysis
summary
System introduces an intermediate language called ArkLang to map underspecified utterances into VizQL, uses syntactic/semantic constraints, heuristics and a relevancy ranking, and implements intra- and inter‑phrasal inferencing to pick attributes, aggregation levels and visual encodings via VizQL. Deployed in Tableau, it collects feedback from 205 users and generates about 6,000 visualizations; evaluations show generally good relevancy but reveal issues with popularity-based attribute choices and a need for personalization and conversational context.
AI-written summary · about
chapters (6)
- 0:00Introduction and motivation
- 2:34Introducing ArkLang and approach
- 7:34Live demonstration of the system
- 13:10Inferencing algorithm details
- 16:22Evaluation and user feedback
- 18:10Summary and future work
related talks
20mEviza: A Natural Language Interface for Visual Analysis
16mType-Directed Synthesis of Visualizations from Natural Language Queries
13mInReAcTable: LLM-powered Interactive Visual Data Story Construction from Tabular Data
20mDataTone: Managing Ambiguity in Natural Language Interfaces for Data Visualization
34mSearching for My Next Chart
42mInteractive Data Analysis: Visualization and Beyond
10mSnowy: Recommending Utterances for Conversational Visual Analysis
46mExplorations in Interactive Visual Analytics