Eviza: A Natural Language Interface for Visual Analysis
summary
A natural language layer sits on an existing visualization and translates utterances into view changes (filter, zoom, select, sort) using a probabilistic grammar refined by the current data, plus semantic enrichment (time, space, units, WolframAlpha/WordNet links). Ambiguous mappings surface as ambiguity widgets (sliders, fuzzy matches) that users adjust, and conversational context (what they call pragmatics) carries intent across queries. A Tableau Research prototype handles spatial/temporal reasoning and a 12-participant study finds NL is faster for focused queries and complements drag‑and‑drop.
AI-written summary · about
chapters (5)
- 0:00Introduction to Eviza
- 1:49Demo of Eviza system
- 7:40Technical contributions overview
- 12:58User study and findings
- 15:47Future improvements and Q&A
related talks
24mInferencing underspecified natural language utterances in visual analysis
20mDataTone: Managing Ambiguity in Natural Language Interfaces for Data Visualization
42mInteractive Data Analysis: Visualization and Beyond
13mInReAcTable: LLM-powered Interactive Visual Data Story Construction from Tabular Data
26mAdding Narrative to BI Dashboards with Natural Language...
10mThe Future of Data Visualization
16mType-Directed Synthesis of Visualizations from Natural Language Queries
01h 00mData, visualization, and designing AI