Constellation: A Visualization Tool for Linguistic Queries from MindNet Tamara Munzner François Guimbretière Stanford University George Robertson Microsoft Research
Overview solve specific problem –help linguists improve MindNet algorithms chosen techniques –custom semantic layout –perceptual channels –interaction as first-class citizen
Definition Graph dictionary entry sentence nodes: word senses links: relation types
Semantic Network definition graphs as building blocks unify shared words large network –millions of nodes –global structure known: dense probes return local info uses –grammar checking, automatic translation
Path Query best N paths between two words words on path itself definition graphs used in computation
Task: Plausibility Checking paths ordered by computed plausibility researcher hand-checks results –high-ranking paths believable? –believable paths high-ranked? –gross polluters (stop words)
Top 10 Paths: kangaroo - tail
Goal create unified view of relationships between paths and definition graphs –shared words are key –thousands of words (not millions) special-purpose algorithm debugging tool –not understand the structure of English
Semantic Layout reflect dataset characteristics path ordering as backbone fill in definition graphs
Semantic Layout “plausibility gradient”
Semantic Layout “plausibility gradient” –horizontal position
Semantic Layout “plausibility gradient” –horizontal position – size
Semantic Layout edge crossings not minimized
Semantic Layout edge crossings not minimized –false attachment solved with interactive selective emphasis
Perceptual Channels redundant combinations –synergy from multiple codings layout gradient –spatial position, word size –quantitative
Perceptual Channels highlighting: visual popout –saturation –brightness –linewidth ordered –although binary
Perceptual Channels highlighting: visual popout –saturation –brightness –linewidth ordered –although binary
Perceptual Channels hue –relation types green: part-of red: is-a cyan: modifier –word types yellow: path green: definition graph blue: leaf –selective (nominal)
Perceptual Channels orientation –relation types axis-aligned: local slanted: long distance –between instances of same word –selective (nominal)
Perceptual Channels enclosure –definition graphs associated with path word –hierarchy
Interaction see video
Video zoom –software vs. video
Semantic Layout Challenges spatial position encodes path ordering –edge crossings not minimized –clutter reduction: interaction, perceptual channels tradeoffs –spatial encoding vs. information density navigation: intelligent zooming –global, intermediate, local
Semantic Layout Challenges navigation intelligent zooming –global path structure overview –intermediate association of path word and definition graphs –local read single definition graph
Color Scheme [Reynolds94] hues –maximally separated on color wheel saturation/brightness –low for unobtrusive, high for emphasis maximal CRT legibility –black text on colored background
Conclusion targeted case study –small user community techniques –encode dataset structure spatially –multiple perceptual channels –interactive selective emphasis, navigation approach broadly applicable
Acknowledgements MSR linguists –Lucy Vanderwende, Bill Dolan, Mo Corston-Oliver iterative design techniques –Mary Czerwinski discussion –Maneesh Agrawala, Pat Hanrahan, Chris Stolte, Terry Winograd funding –Microsoft Graduate Research Fellowship, Interval Research – –