My homework ! Correlation does not imply causality Visualization for - PDF document

My homework ! “Correlation does not imply causality” Visualization for the Blind ! Or, “ post hoc ergo propter hoc ” ! “after this therefore because of this” ! Sorry about running out of time on Richard Anderson, space shuttle graphs Ken Fishkin, Songyun Hahn Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 Administrivia Administrivia(2) ! Keep working towards your project ! Have you had a conference or journal proposals submission reviewed? ! I’ve already heard 4 – way to go! ! Have you been a reviewer? Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 We’ve seen many visualization techniques But what if you’re blind? Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 1

What about this? What about this? Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 You get the idea Another motivation ! 1.3 million legally blind in US ! “But even well-meaning Web site builders ask: How can I justify the extra ! 68% unemployment rate cost for such a small percentage of the ! WWW has been a negative public?” (Steven Pemberton, ACM ! Criminally overlooked area Interactions, Feb 2003, p. 44) Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 Another motivation Mynatt paper ! “The answer is: Google. … It is clear ! Really two papers in one – presents that at least half the visitors find the both “Mercator” and “GUIB”. sites via Google. And what Google sees ! Two different systems for providing is exactly what a blind person sees. access to graphical applications Google is a blind user – a billionaire blind user, with millions of friends who ! Fairly old (predates Web), foundational listen to its every word. If a blind user can’t see your site, neither can Google, and you site will suffer.” (Pemberton) Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 2

Mercator vs. GUIB Goal: Coherence with Visual ! Both assume app is “black box” ! “An overriding concern … is maintaining ! GUIB relies on tactile output – takes coherent, parallel … interfaces”. Why? screen pixels and maps them to a 2D ! Collaboration (“primary reason”) dynamic braille display ! “to support discussion” ! Mercator relies on audio output – takes ! “to support simultaneous interaction” screen objects and maps them to audio. ! Training ! (some of this is due to Europe vs. US) ! 4 design goals Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 Goal: convey visual Goal: Information packing information ! Need to support random-access of an ! Not enough to just translate words information-rich environment (e.g. ! 1 step up: buttons, menus mouse-click anywhere on a map) ! Next step up: sliders, scrollbars ! Uh-oh: random graphics Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 Goal: Interaction Goal: wide scope ! Treat apps as “black boxes”, so can ! Can only rely on keyboard, and audio handle multiple apps output (not even audio input). ! Both monitor GUI painting calls and ! How do you do WIMP without P? build a model of the screen ! GUIB in Windows/X ! Mercator in X ! Imposes a severe limit on how smart they can be Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 3

GUIB GUIB: text ! “GUIB continues the use of the spatial ! Caret is a conceptual object, but is metaphor” – a line is a line. mapped physically ! Map is maintained from NV display to V display, on a per-pixel basis. Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 GUIB: window GUIB: icon ! User feels the menu bar Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 GUIB: menu GUIB: scroll bar ! Note how selection is shown ! works well here Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 4

GUIB: list box GUIB: button Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 Summary: Mercator ! + maps a 2D display to a 2D display, ! Instead of 2D display, uses sound can do it “all at once” ! Sound is 1D, what do you do? ! - adds junk user doesn’t care about ! Have a focus at any given point, ! - requires 2D braille display describe what’s around the focus ! Use other dimensions of sound (pitch, “earcons” to annotate) Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 Mercator: text Mercator: window ! Uses pitch, “earcon” to show ! Uses earcons – you get the idea attributes/context Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 5

Mercator: icon Mercator: menu Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 Mercator: scroll Mercator: list box ! Requires fine pitch discrimination Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 Mercator: button Questioning the assumptions ! Should Visual/Non-Visual UIs necessarily have a 1-1 map in concepts and metaphors? ! For example, GUIB approach is rather like “curses”: Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 6

Counter-Example Counter-Example: ! Nethack, 2D: Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 Nethack vs. Diablo Yu paper ! They didn’t keep the same metaphors, ! Previous works focused on (largely) textual data, in a 2D why should we? arrangement. What about visualizations? ! Multivis (http://www.multivis.org) project looking at bar charts, graphs, pie charts. ! You were assigned 1 paper of a “suite” Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 Basic concepts Haptic ! Like earlier papers, use haptic or audio ! Use (a) “phantom” ! Go beyond them by combining them (3D force- ! Ditch “Coherency” constraint feedback bat) ! Semi-real user studies to test (more on (http://www.sensable.com/h ! aptics/products/phantom.htm this later) l) Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 7

Haptic Haptic ! Or more commonly ! Make a “sculpture” out of visualization, with Logitech force- “grooves” felt by force-feedback. feedback mouse ($60 vs $10K) Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 Line graph Bar Chart Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 Pie Graph Note ! 3 rd dimension not really used with phantom ! Also mentioned that felt limited, you mainly “poke” to sense the outline of the shape. Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 8

Audio Non-speech vs. speech ! Line: MIDI, high Y " high pitch ! In later paper, also tried speech sounds (voice “reads” value on graph) ! -- assumes lines with unique Y for each X ! Found non-speech significantly better ! Bar: MIDI, high Y " high pitch ! Later work: high X " right ear ! Pie: MIDI, high % of pie " high pitch ! In all cases, sound starts/stop when region entered/left Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 They tested it! Digression ! Most tests were of undergrads wearing ! Example of why user testing is hard, blindfolds and rarely done, or done well ! Particular issue in this environment ! Hard to find users ! Human subjects approval takes a long time Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 Digression(2) More on the testing ! Blindfolding questionable: very different ! To be rigorous in the testing: set of haptic/audio skills, different ! 3 types of graphs expectations, etc. ! * 4 conditions = 12 scenarios ! However, they did do some testing with ! * n >= 5 (at least) blind as well. ! " 60 people ! Surprisingly (to me), found only quantitative, not qualitative, differences in the two user groups Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 9

The Results This just in ! Haptic is better than nothing ! Recent Roth/Kamel/Petrucci/Pun paper (“A Comparison of Three Nonvisual Methods for Presenting Scientific Graphs”, J. ! Little things make a big difference: Visual Impairment and Blindness, June 2002, 96 (6)) found similar separating bars hurts results: ! Audio is better than haptic ! Haptic + Audio better then either ! Haptic + Audio is better than either ! Presenting continuous graphs ! Did real user testing: sighted and blind had similar qualitative results Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 Background Automatic tactilization of ! Graphical representation for the blind graphical images - Audio description - Braille graphics - Brief introduction of the project - Haptics based on NSF proposal - - Braille is the best modality for image comprehenssion (Skiff, 2002) January 22, 2003 Sangyun Hahn Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 Production process of braille Tiger embosser graphics ! The first braille printer that supports Windows-style WYSIWYG printing ! Can print braille characters and graphics together ! Print with highest resolution (20 dpi) among currently available embossers ! Can emboss dots with different heights. Jan 22, 2003 CSE 510 - Winter 2003 Jan 22, 2003 CSE 510 - Winter 2003 10