Efficiency or Quality of Experience: A Laboratory Study of Three Eyes- Free Touchscreen Menu Browsing User Interfaces for Mobile Phones

The Open Rehabilitation Journal, 2011, 4, 13-22 13

Open Access
Efficiency or Quality of Experience: A Laboratory Study of Three Eyes-
Free Touchscreen Menu Browsing User Interfaces for Mobile Phones
Vladimir Kulyukin*,1, William Crandall2 and Daniel Coster3

1
Department of Computer Science, Utah State University, Logan, UT, USA
2
Smith-Kettlewell Eye Research Institute, San Francisco, CA, USA
3
Department of Mathematics and Statistics, Utah State University, Logan, UT, USA

Abstract: A growing number of individuals who are blind or visually impaired is using smartphones in their daily
activities. The touchscreen is a standard component of smartphones. While benefitting people with low vision by
enhancing control of the text style and color and the size of images and text, the touchscreen has the downside for visually
impaired users in that physical buttons for input of command selection and text entry are replaced with the touchscreen’s
soft buttons. To overcome this limitation, we are investigating eyes-free approaches to using the smartphone’s
touchscreen for information browsing. In this article, we present a laboratory study of three eyes-free touchscreen user
interfaces for browsing menu hierarchies. Our findings indicate that quality of experience and familiarity may be as
important as the time efficiency of completing tasks.
Keywords: Mobile computing, eyes-free computing, touchscreen gesture interfaces, blindness, visual impairment.

1. INTRODUCTION the touchscreen’s soft buttons. Although replacing the mouse
by voicing the area of finger contact on the screen is
A growing number of individuals who are blind or
available and shortcuts in typing through word completion
visually impaired (VI) are using smartphones in their daily are quite effective in reducing the number of keystrokes
activities [1]. Smartphones provide an impressive
necessary to perform a text entry task, real time manipulation
compliment of features in a compact, portable form factor
of these interface options are inefficient in the context of
suitable for executing or contributing to real-time
using the smart phone for wayfinding when on the street.
wayfinding tasks. GPS with Geographical Information
Furthermore, the current convention in web access is to drill
System (GIS) map data is included with most, if not all,
down the hypertext-linked, hierarchical menus where text
smart phones. The “open” architecture of the Android entry is minimized and rapid scanning of text at each level is
operating system allows inexpensive, special purpose
desirable.
applications (“apps”) to be readily designed and distributed.
In addition to two-way voice and data communication, many Toward this end, we are investigating eyes-free
provide effective text-to-speech (TTS) and moderately approaches to using the smartphone’s touchscreen and
effective speech-to-text (STT) functionality as user interface trackball for gesture control over information navigation,
(UI) options. thus giving blind and VI users access to information that is
either stored on the smartphone or brought into the
The touchscreen is a standard component of
smartphone through its wireless connection. Such systems
smartphones. Manufacturers prefer “soft” controls (for such
are useful in a variety of contexts ranging from accessible
functions as keyboards and pushbuttons) to hardware ones as
shopping [2] to remote infrared audio signage (RIAS) [3] to
the touchscreen replaces expensive and less reliable
indoor and outdoor navigation [4].
mechanical parts and reduces the device footprint.
Consumers, for the most part, find touchscreens efficient. Our approach complements and draws on previous and
Therefore, one should expect to see relatively more current research on touchscreen accessibility. The Slide Rule
touchscreens on phones with the possibility that mechanical interface [5] provides several accessible multi-touch
controls may disappear altogether. While benefitting people interaction techniques for touchscreen interfaces for
with low vision by enhancing control of the text style and browsing lists, selecting items, and browsing hierarchical
color and the size of images and text, the touchscreen has the information. The EarPod system [6] provides access to
downside for VI users in that physical buttons for input of hierarchical audio menus through a circular touchpad. The
command selection and text entry have been replaced with Talking Fingertip technique [7] allows blind and VI users to
scan touchscreens and hear the descriptions of the items on
the screen. In the Talking Tactile Tablet [8], a stylus can be
*Address correspondence to this author at the Department of Computer used to explore two-dimensional space and receive feedback
Science, Utah State University, Logan, UT, USA; through speech and a tactile overlay. The Touch 'n Talk
Tel: (435) 797-8163; Fax: (435) 797-3265;
E-mails: vladimir.kulyukin@usu.edu, vladimir.kulyukin@gmail.com
system uses speech and tactile overlays to enable the users to
edit text documents. As an alternative to the touchscreen, the

1874-9437/11 2011 Bentham Open

14 The Open Rehabilitation Journal, 2011, Volume 4 Kulyukin et al.

BlindSight system [9] uses the phone keypad to access described in Section 5. Section 6 gives our results. In Section
speech menus. We endorse the research objectives of these 7, the results are discussed. A summary and conclusions are
approaches and contribute to them by first focusing on the presented in Section 8.
efficacy of simple, single finger methods appropriate to
browsing information while ‘on the go’. 2. EYES-FREE USER INTERFACES

We took the measure of the complexity of a browsing To evaluate the suitability of touchscreen interfaces for
task to be the minimum number of UI actions required to eyes-free information browsing on smartphones, we have
transverse a tree in order to reach the answer to a particular designed and implemented three UIs. All three UIs are based
query. By evaluating subjects’ behavior in solving various on the common DPAD metaphor (up, down, left, right, select
eyes-free search navigation problems using three modes of gestures) familiar to many smartphone users. The UIs are
finger gestures (UIs), we provide a useful model for designed for browsing single-inheritance hierarchies, i.e.
evaluating other information browsing methodologies. The trees where each node, except the root, has exactly one
project was implemented on the Google Nexus One parent and the root node has no parent (See Fig. 1). This type
smartphone running Android 2.2. The Android OS was of information organization is very common and can encode
selected to maximize potential public impact. Unlike the many semantic relationships from standard menu hierarchies
proprietary iPhone OS, Android is open source, can be to sophisticated XML ontologies. We make no claims about
programmed with standard programming languages (Java, the generalizability of our interfaces to other knowledge
C/C++), has emulators and integrated development encoding structures. A standard menu layout of the tree
environments (IDEs) for Windows, Linux, and Mac OS, structure is assumed, with the root being the leftmost node
requires no mandatory developer fees, which is fundamental and the leaves being the rightmost nodes, as shown in Fig.
to scientific knowledge sharing, and is gaining momentum in (1). When the focus moves to a specific node, the node is
the mobile phone accessibility literature [10]. We hope that said to be activated. An activated node may be rendered
the end product will be an open source solution for all through audio, vibration or, for low-vision users, even
Android platforms such as Motorola BackFlip, Sprint’s HTC graphically.
EVO, Google Nexus One, Google Droid, and Google Droid 2.1. User Interface 1
Incredible).
The first UI (UI1) is a joystick-type interface with five
This remainder of the article is organized as follows. In user operations: up, down, left, right, select. UI1 was
Section 3, each UI is described in detail. The browsing tasks implemented with the Google Nexus One trackball (small
and methodology of measuring their complexity are round ball at the bottom of the phone). At each activated
presented in Section 4. The design of our experiments is node in the tree, the user can execute five actions:

Fig. (1). Sample Menu Hierarchy.

Efficiency or Quality of Experience The Open Rehabilitation Journal, 2011, Volume 4 15

1. trackball-left (move left to the parent of the current 2.2. User Interface 2
node and hear result);
The second UI (UI2) is a one finger touch gesture
2. trackball-right (move right to the first child of the interface with five gestures:
current node and hear result);
6. finger-left (user moves finger left on touch screen);
3. trackball-up (move up to the previous node on the
same level and hear result); 7. finger-right (user moves finger right on touch
screen);
4. trackball-down (move down to the next node on the
same level and hear result); 8. finger-up (user moves finger up on touch screen);
5. trackball-tap (select current node). 9. finger-down (user moves finger down on touch
screen);
For an example of how UI1 works, consider the sample
hierarchy in Fig. (1). Suppose the user starts at the root node 10. finger-tap (user taps finger on touch screen).
Zeum Art and Technology Center and hears the spoken Suppose the user again starts at the root and hears “Zeum
message “Zeum Art and Technology Center” through the Art and Technology Center.” The UI2 actions and their
TTS engine. If the user executes trackball-left, the user effects are summarized in Table 3.
hears the spoken message “No parent node,” because the
root node has no parent. If the user executes trackball-right, Suppose the user executes finger-right to move to Hours
the focus moves right to the node Hours of Operation, which of Operation. The UI2 actions and their effects when Hours
becomes activated, and the user hears the spoken message of Operation is activated are given in Table 4.
“Hours of Operation.” If the user executes trackball-up, the To implement these gestures, we divided the Nexus One
user hears “No previous node,” because there is no node touch screen into four quadrants: 1, 2, 3, and 4, starting with
upward of Zeum Art and Technology Center on the same 1 as the top left quadrant and moving clockwise. A touch
level of the hierarchy. If the user executes trackball-down, gesture is classified as finger-left if the gesture starts in
the user hears “No next node” as there is no node below the quadrants 2 or 3, ends in 1 or 4, and the line passing through
root node. If the user executes trackball-tap, a specific the start and end points has the absolute slope of no more
action associated with Zeum Art and Technology is executed than 45 degrees. A touch screen is classified as finger-right if
by the system. In our current implementation, the user hears it starts in 1 or 4, ends in 2 or 3, and the absolute slope of the
the spoken message consisting of the text associated with the line through the start and end points does not exceed 45
node in a database. These UI1 actions and their effects are degrees. A gesture is classified as finger-down, if it starts in
summarized in Table 1. 1 or 2, ends in 3 or 4, and the line's absolute slope is no more

Table 1. UI1 Actions and Effects at Zeum Art and Technology Center Node in Fig. (1)

UI1 Actions Effects

trackball-left Message “No parent node.”
trackball-right Hours of Operation is activated; Message “Hours of Operation.”
trackball-up Message “No previous node.”
trackball-down Message “No next node.”
trackball-tap Action associated with node Zeum Art and Technology Center is executed.

Suppose, to continue with our example, the user is at the than 45 degrees. A gesture is classified as finger-up, if it
root node and executes trackball-right to move to Hours of starts in 3 or 4, ends in 1 or 2, and the line's absolute slope is
Operation. The UI1 actions at Hours of Operation and their no more than 45 degrees. Finally, a gesture is classified as
effects are summarized in Table 2. finger-tap if the size of the segment that connects the start
and end point of the gesture is below a specific threshold.

Table 2. UI1 Actions and their Effects at Hours of Operation in Fig. (1)

UI1 Actions Effects

trackball-left Zeum Art and Technology Center is activated; Message “Zeum Art and Technology Center.”
trackball-right School Year Hours is activated; Message “School Year Hours.”
trackball-up Message “No previous node.”
trackball-down Fees is activated; Message “Fees.”
trackball-tap Action associated with Hours of Operation is executed.


Table 3. UI2 Actions and their Effects at Zeum Art and Technology Center in Fig. (1)

UI2 Actions Effects

finger-left Message “No parent node.”
finger-right Hours of Operation is activated; Message “Hours of Operation.”
finger-up Message “No previous node.”
finger-down Message “No next node.”
finger-tap Action associated with node Zeum Art and Technology Center is executed.

Table 4. UI2 Actions and their Effects at Hours of Operation in Fig. (1)

UI2 Operations Action Effects

finger-left Zeum Art and Technology Center is activated; Message “Zeum Art and Technology Center.”
finger-right School Year Hours is activated; Message “School Year Hours.”
finger-up Message “No previous node.”
finger-down Fees is activated; Message “Fees.”
finger-tap Action associated with Hours of Operation is executed.

Fig. (2). A sequence of UI3 screen shots.

2.3. User Interface 3 screen, the user hears “Hours of Operation” spoken through
TalkBack. The user uses the trackball to go up and down the
User Interface 3 (UI3) is based on three accessibility list. When the user goes down the list (screen shot 4 in Fig.
features natively available on the Android OS: TalkBack, 2), the user hears “Fees” accompanied by a beep and a light
SoundBack, and KickBack [11]. This UI may be viewed as a vibration. When the menu item Fees is selected, the screen
reasonable benchmark against which other methods can be displays the menu shown in screen shot 5 and hears the
compared. TalkBack reads currently highlighted items on the message “Adults seven dollars.”
screen; SoundBack beeps when the user does something with
the screen; KickBack is similar to SoundBack but uses 3. BROWSING TASKS
gentle vibrations instead of beeps. In UI3, all three features
We define a browsing task as an information query that
were turned on. The user is given the choice of using either
the user must answer by browsing a specific menu. For
the trackball or a finger for navigation and selection.
example, in the menu in Fig. (2), the user may want to know
Fig. (2) gives an example of how UI3 works. Suppose the fee for the children under four. The complexity of a
that the user wants to find the admission fee for adults for the browsing task can be measured by the number of available
Zeum and Art Technology. Moving from left to right, the UI actions taken by the user to find an answer to the query.
first picture in Fig. (3) is the first screen shot. The user uses Since we are browsing single-inheritance hierarchies, the
the trackball to select the only item on the screen. As soon as complexity of a browsing task can be measured in terms of
this item is selected (screen shot 2 in Fig. 2), the user hears five abstract user actions: up, down, left, right, and select.
the name of the item through TalkBack, hears a beep through Each of the three UIs offers a specific implementation of
SoundBack, and feels a light vibration through KickBack. each abstract action. For example, in UI1, the up action is
The phone screen now displays the menu shown in the third implemented with trackball-up, the down action is
screen shot in Fig. (2). When this menu appears on the implemented with trackball-down, etc.


To standardize task complexity measurement, we assign hierarchy if there is no other query for that hierarchy that can
a number to each browsing task that measures the smallest be answered in a larger number of abstract actions.
number of actions required to complete it. This number
To illustrate these definitions, consider again the menu
approximates the number of actions that a knowledgable user
hierarchy given in Fig. (2). The query “What are the school
(the one who knows where the answer is in the menu) will year hours of operation Wednesday through Friday?” is easy,
execute to find the answer. Suppose again that the user wants
because it can be answered in four actions, which is the
to find an answer to the query is “What is the fee for the
smallest for the hierarchy. On the other hand, the query
children under four?” in the menu in Fig. (3). The smallest
“What is the fee for the children under four?” is hard,
number of abstract actions from the root node Zeum Art and
because it can be answered in six actions, which is the
Technology Center to the node Free is six: 1) right (to Hours
largest number of actions for the hierarchy. It should be
of Operations), 2) down (to Fees), 3) right (to Adults), 4) noted that, under these definitions, a hierarchy may have
down (to Youth), 5) down (to Children Under Four), and 6)
several easy and hard queries.
right (to Free). Thus, the complexity of this browsing task in
this menu is six. 4.3. Design and Method
Under this measurement scheme, browsing tasks can be Participants
meaningfully compared with each other to determine if one
is harder than the other. For example, if we compare the Seven visually impaired participants were recruited for
browsing task “What are the Sunday hours of operation the experiments. The age ranged from thirty two to sixty.
during the school year?” with the browsing task “What is the Five were completely blind; one had some peripheral vision.
fee for the children under four?”, our complexity None of the subjects was able to read the mobile phone
measurement method ranks the former task as easier than the display. There were two guide dog users and four cane users.
latter, because the former task's complexity of four (right, Two subjects never used a mobile phone and never had one;
right, down, down) is smaller than the latter task's the other five had mobile phones and used them for phone
complexity of six. calls.
Participant Training
4. EXPERIMENTS
Each session was broken into three parts. The first part
4.1. Training and Experimental Menus
was a tutorial. Each participant was instructed on how to use
We have designed five menus for our experiments: two each of the three user interfaces. The participant was given a
for user training and three for the actual experiments. The Google Nexus One smartphone with the software installed
first training menu consisted of thirteen menu items that on it and was shown how to use each interface on two
varied from letters to numbers to color names and had three training hierarchies. When the participant was comfortable
levels. The second training menu consisted of twenty five with a specific interface, he or she was given two training
nodes and had four levels. The second training menu coded a browsing tasks to complete. The training was finished when
taxonomy of plants (trees, bushes, and flowers) in various the participant successfully completed both browsing tasks.
geographic regions of the U.S. While completing the training tasks the participant was
allowed to ask the experimenter for assistance.
The three menus used in our tests contained information
on the Yerba Buena Gardens, a public recreational facility in Experimental Methods
San Francisco, CA. We chose to represent information about
Once trained, each participant was asked to use each of
this facility, because it already has several Talking Signs
the three user interfaces to complete one hard browsing task
transmitters installed on its premises. Therefore, in the
future, when the hardware component of our system is ready, on each of the three experimental menus. Subsequently,
three participants, randomly chosen, were asked to complete
we hope to test it on the premises. The first experimental
an easy browsing task (in addition to the hard one) on each
menu encoded information about the Yerba Buena Skating
of the three experimental menus. Availability of participants
rink. It had twenty three nodes and four levels. The second
and time and budget constraints on the study dictated that
experimental menu encoded information about the Yerba
only three participants completed this phase of the study. In
Buena Rocco's Restaurant. It contained fifty nodes and had
four levels. The third experimental menu encoded the third phase, the participants were given an opportunity to
provide feedback on each interface.
information about the Yerba Buena Zeum and Art
Technology Center. It contained eleven nodes and four To estimate the potential of speech-based information
levels. browsing, five participants were asked to speak a series of
twelve one or two word commands to the speech-to-text
4.2. Easy and Hard Queries engine of the Android Incredible smartphone. The
For each test hierarchy, we designed two queries: one commands are listed in Table 9. For each command the
easy and one hard. Given a hierarchy, an easy query is a number of repetitions was recorded before the command was
query with the smallest browsing task complexity, as recognized by the engine. If a command was not recognized
measured by the method described above, whereas a hard after five repetitions, the command was considered
query is a query with the largest complexity. In other words, unrecognized and scored as a 5. Thus, for each command,
a query is easy for a given hierarchy if there is no other the number of repetitions ranged from 1 to 5.
query for that hierarchy that can be answered in a smaller A qualitative survey was taken of each participant after
number of actions. Similarly, a query is hard for a given the formal experiments. Each participant was asked to


specify which UI the participant preferred and why. The Statistical Power and Effect Sizes
participants were also given the opportunity to offer free-
Observed effect sizes (Cohen’s d, mean difference divided
form comments on touch gesture interfaces and their
by estimated standard deviation) are provided in Tables 6 and 8.
experience with them. The participants’ responses were
The small n-size of this study, 7 participants for the primary
logged into text files.
phase of data collection, dictated that only large effect sizes
Statistical Design greater than (about) 1 could be easily detected. The approximate
power to detect an effect size of 1 (at least two means differing
To minimize bias due to a potential learning effect, the
by 1 standard deviation) was ~87%. Consequently, only large
order in which each participant used each of the three user
effects were found to be statistically significant in this study.
interfaces was randomized for each participant, as was the
Because of the exploratory nature of this study, no adjustment
assignment of a particular menu query (easy or hard, as
appropriate) for each participant using each interface. Since was made to P-values for multiple comparisons.
each participant used each interface, the experiment was Data Collection Protocols
formally a repeated measures (specifically, a cross-over)
As the participants used the software, all their actions were
experiment with user interface the “within subjects” factor of
automatically logged and timestamped by a program running on
primary interest. The effects of gender and any previous
the smartphone. When a participant forgot what he or she was
mobile-phone experience were controlled for as “between
subjects” factors when analyzing the results for all seven looking for or became completely lost in the menu, the
participant was allowed to stop the experiment and ask for
participants. For the three participants who answered both
clarification or ask to start over. All participants were
easy and hard queries with each interface, a second factor,
encouraged to do the best they could.
“EasyHard” with two levels, was included along with its
interaction with user interface. 5. Results
Variables 5.1. Hard Queries
For the experimental data, the measured dependent The first repeated measures ANOVA analysis modeled the
variable was completion time. The log of the completion data collected on the seven participants who were asked to use
time was analyzed to reduce the adverse impact of non- each of the three UIs to answer a hard query on each of the three
normality due to observed right skewness of the raw experimental menus.
completion time data. The independent variable was factor
(UIID, user interface ID at three levels: UI1, UI2, and UI3, Overall, adjusting for gender and any previous mobile-
each corresponding to the specific UI. Since one hard query phone experience, mean log-completion time differed
was randomly assigned to each of the three UIs and each of significantly among the three user interfaces, (F(2,16) = 6.04, P
the three experimental menus, there were a total of twenty = 0.0111). Table 5 below summarizes the analysis and shows
one observations in the primary phase of the study with each that UI3 had the lowest observed mean log-completion time
of the seven participants completing three runs, each run with all seven participants using one randomly assigned hard
consisting of a user interface coupled with a randomly query for each of the experimental menus. The last column of
assigned query. Other independent factors were gender (2 Table 5 converts means back to the measured time scale
levels) and any previous cell-phone experience (2 levels). (seconds) for ease of comparison.

Statistical Model Table 6 provides the t-statistics, P-values, and observed
effect sizes for each user interface comparison. Table 6 shows
After transformation to the log-scale, residual analysis that the mean log completion time for UI3 was significantly
indicated approximate normality (Shapiro-Wilk test, P > lower than the mean log completion times for both UI1 (t =
0.95), no more than modest non-constant variance, and no 2.391, P = 0.030) and UI2 (t = 3.375, P = 0.004). The mean log
indication of auto-correlation within each participant’s three completion times for UI1 and UI2 did not differ significantly.
measurements (i.e., no significant autocorrelation within Thus, UI3 is the clear winner in this analysis. UI2 was the least
participants was found using SAS PROC MIXED correlation efficient, on average, but not significantly worse than UI1. The
structures). Additionally, no major outliers were observed. third entry in each cell of Table 6 is the observed effect size:
Consequently, a standard parametric repeated measures mean difference divided by the (estimated) standard deviation.
ANOVA was used to assess differences in mean log
completion time among the three user interfaces, with 5.2. Easy vs Hard Queries
adjustment for gender or previous mobile phone experience Three of the seven participants were randomly selected to
as appropriate. All analyses were conducted using the SAS answer both easy and hard queries on the assigned menus.
statistical software [13]. Thus, in addition to the hard queries, these individuals were

Table 5. Mean Log Completion Time (seconds) for Three UIs

UUID Mean Completion Time (Log Scale) Standard Error (Log scale) Mean Completion Time (sec)

UI1 5.461 0.241 235
UI2 5.794 0.241 328
UI3 4.654 0.241 105


given one easy query each and therefore this part of our also significantly less efficient than UI3, although only
experiment produced eighteen observations (each participant marginally so (P = 0.0475). Finally, for UI3, there was no
completing six runs). The independent factors were again significant difference in mean log completion time (P =
User Interface ID (UIID) and EasyHard (coded 1 for Easy 0.2169), but hard queries took significantly longer than easy
and 2 for Hard), with Gender as an adjustment effect. The queries for both the other interfaces: P = 0.0112 for UI1 and
dependent variable was the log of completion time. P = 0.0001 for UI2. Mean log completion time was
Table 6. t-Statistics and P-Values and Observed Effect Sizes
significantly greater for hard queries with UI2 than any other
when Comparing Mean Log Completion Time interface or query difficulty (see significant P-values in
column label UI2 Hard in Table 8).
t-Statistic/ P-Value/ Effect Size Comparing Mean of UIi to UIj For hard queries, UI3 is the clear winner, while UI2 the
worst. For easy queries, the choice of interface does not
UI2 UI3 matter, although UI3 had the lowest observed mean
completion time. Thus, for time efficiency without knowing
-0.989 2.391
UI1 0.337 0.030
the difficulty of the query, UI3 is the UI to use.
0.530 1.281 After the experiments were completed, we asked each
3.375 subject to decide on their preferred interface. Five subjects
UI2 0.004 chose UI2, two subjects chose UI1. None chose UI3. In other
1.812 words, most subjects preferred the least time-efficient user
interface over the two more time-efficient counterparts.
The overall ANOVA model was significant, F(6,11) = As a pilot for future work, we looked briefly at speech
19.14, P < 0.0001. F-tests for each factor’s effects are recognition of simple commands relevant to the browsing
presented below in Table 7. All effects were significant at task under consideration. Seven out of twelve speech
the 0.05 level, including the interaction of UIID with commands were recognized in fewer than two repetitions
EasyHard, adjusting for Gender. (Table 9).
An interaction line-plot for the mean log completion time 6. DISCUSSION
for UUID and EasyHard queries at each user interface is in
Our experiments show that, for our sample, UI3 was the
Fig. (3) below. The lower blue line plots the mean log
clear winner in terms of time-efficiency of task completion,
completion times of the easy queries against the three user
while UI2 was the worst. When the participants needed to
interfaces. The upper dashed curve plots the mean log
complete easy queries, the choice of interface did not matter,
completion times of the hard queries for the three user
although UI3 had the lowest observed mean completion
interfaces. The significant interaction between UIID and
time. The qualitative survey of the participants showed that
EasyHard is visually evident in that the blue and red lines are
most participants (five out of seven) preferred the UI2
non-parallel. Specifically, the magnitude of the difference
interface, with the two other subjects opting for UI1. None
between the mean log completion times for an easy versus
chose UI3. In other words, the participants preferred the two
hard query depends of the user interface. The difference is
less time-efficient user interfaces over the most time-
large for UI2, moderate for UI1 and small for UI3. It is also
efficient interface.
visually apparent that mean log completion time was
effectively the same for Easy queries for all UIs (blue line is This incongruence between quantity and quality suggests
nearly flat), but this was not the case for Hard queries, where that time-efficiency was not the only factor that determines
UI2 was less efficient than either UI1 or UI3. the user preference. Since six out of seven participants in our
sample used mobile phones on a regular basis, they quickly
Table 8 provides the detailed assessment of the
became comfortable with the five gesture DPAD metaphor
differences in mean log completion time among the six
and learned to browse the menus with the five touch screen
combinations of UIID and EasyHard. As suggested by the
gestures. The two users who preferred the trackball interface,
blue line in Fig. (3), there was no significant difference in
UI2, said that they preferred keys to gestures, because they
mean log completion time among the three User Interfaces
had phones with a physical keypad. The touchscreen was
for easy queries: UI1 versus UI2, P = 0.4594; UI1 versus
completely new to them and they said that they needed more
UI3, P = 0.4347; and UI2 versus UI3, P = 0.2317. In
time to become comfortable with it.
contrast, for hard queries, UI2 was significantly less efficient
than both UI1 (P = 0.0060) and UI3 (P = 0.0002). UI1 was If the participants preferred UI2, why did they fail to
complete the queries with UI2 as fast or faster than with UI1

Table 7. F-Tests for Effects of UIID and EasyHard

Source DF Sum of Squares Mean Square F Value Pr > F

UIID 2 3.139 1.569 12.03 0.0017
EasyHard 1 4.363 4.363 33.45 0.0001
UIID*EasyHard 2 1.255 0.627 4.81 0.0315
Gender 1 6.221 6.221 47.70 <.0001


Fig. (3). Interaction Plot of UUID and EasyHard.
Table 8. t-Statistics, P-Values and Effect Sizes for Combinations of UIID and EasyHard

t – Statistic for Comparing Each UI and EasyHard Mean/P_Value/Observed Effect Size
i/j UI1 Hard UI2 Easy UI2 Hard UI3 Easy UI3 Hard

-3.04 -0.77 -6.43 0.50 -0.811
UI1 Easy 0.011 0.459 <.001 0.628 0.435
2.59 0.64 5.36 0.42 0.66
2.27 -3.39 3.54 2.23
UI1 Hard 0.044 0.006 0.005 0.048
1.88 2.77 2.91 1.83
-5.67 1.27 -0.04
UI2 Easy <0.001 0.232 0.965
4.63 1.02 0.03
6.93 5.62
UI2 Hard <.001 <0.001
5.63 5.69
-1.31
UI3 Easy 0.217
0.60

or UI3? One conjecture, based upon our observations of the slower browsing times for UI2. Although the time to
training sessions, is that there may simply be a semantic navigate these test menus are in excess of what is practical
dissonance between the UI2 gestures and how many people for “on the go” use, further research will help clarify the
understand the term “hierarchy.” Many people understand above conjecture about semantics and possibly lead to an
this term as a genealogical tree where a child node resides overall increase in the time-efficiency for these methods. In
below its parent node. Thus, one has to go “down” to get future work, we will also test other accessible information
from a parent to a child and go “up” to get from a child to a browsing systems such as the IDEAL Web Reader, a
parent. In UI2, on the other hand, one must go right to go gesture-based open source web browser (www.apps4andro
from a parent to a child and to go left from a child to a id.org).
parent. This cognitive mismatch might have caused the


Table 9. Speech Commands and Average Number That semantic understanding may simply not exist if the
Repetitions for Recognition Required computer literacy levels of the target population vary
significantly. For example, two of our participants who use
Commands Num. Repetitions touch screen phones on a regular basis had no problem
understanding the menu and DPAD metaphors on which UI2
1. left 1.2 is built.
2. right 1 No far reaching conclusions can be drawn from our,
rather informal, evaluation of speech commands. A high
3. up 1.8
recognition rate across the speakers is testimony to the
4. down 1 quality of Google's web-based speech recognition service
5. press 1.4 accessible through an application on the
6. select 1.6 Android Incredible smarthphone. It is reasonable to
expect that that the quality of mobile speech recognition will
7. move left 2.8 improve even more in the future and will handle a greater
8. move right 2.6 variety of accents. Thus, speech may become a viable option
for controlling eyes-free information browsing. One caveat,
9. move up 1.8
however, is that, even if speech recognition becomes
10. move down 2.6 completely accurate, some people may still have privacy
11. item press 4.2 reservations of speaking commands to the system in public
spaces [12].
12. item select 2.8
The efficiency of UI3, based on the default accessibility
options available on the Google Nexus One phone, testify to
Regarding the pilot for speech input, in addition to the the fact that the accessibility of the Android platform is
small number of participants - all of whom were native improving [10]. The fact that none of the participants chose
English speakers - it should be noted that the speech UI3 as the interface that they would use indicates that quality
recognition trials were done under laboratory conditions with of experience and familiarity may be as important as the time
no extraneous sounds present. Since real world environments efficiency of completing tasks.
have substantial background noise, the performance of the
ACKNOWLEDGMENTS
present speech-based browsing task would certainly be
degraded. Speech input for browsing control may play a We are grateful to the participants for their productive
bigger role as reliable operation in noise increases and as and constructive feedback. This research has been supported,
natural language processing develops. in part, by The Smith-Kettlewell Eye Research Institute and
the National Institute on Disability and Rehabilitation
9. SUMMARY AND CONCLUSIONS Research to the Smith-Kettlewell Rehabilitation Enginee-
We designed and implemented three eyes-free UIs for ring Research Center on Low vision and Blindness, San
browsing single-inheritance hierarchies. We took the Francisco, CA.
measure of the complexity of a browsing task to be the
REFERENCES
minimum number of UI actions required to transverse a tree
in order to reach the answer to a particular query. Seven VI [1] Kientz J, Patel S, Tyebkhan A, Gane B, Wiley J, Abowd G.
participants were each given a Google Nexus One Where’s My Stuff? Design and evaluation of a mobile system for
locating lost items for the visually impaired. In: Proceedings of the
smartphone (with the UI software installed) for a hands-on 8th ACM Conference on Computers and Accessibility (ASSETS
demonstration of each of the three interfaces on two training 2006), Portland, Oregon, USA, October 22-5, 2006.
hierarchies. Three participants, randomly chosen, were asked [2] Kulyukin V, Kutiyanawala A. From ShopTalk to ShopMobile:
to complete an easy browsing task (in addition to the hard vision-based barcode scanning with mobile phones for independent
blind grocery shopping. In: Proceedings of the 2010 Rehabilitation
one) on each of the three experimental menus. For each test Engineering and Assistive Technology Society of North America
hierarchy, we designed two queries: one easy and one hard. Conference (RESNA 2010), Las Vegas, NV 2010.
The participants were given an opportunity to provide [3] Crandall WF, Brabyn JA, Bentzen B, Myers L. Remote infrared
feedback on each UI. The experiments showed that UI3 (the signage evaluation for transit stations and intersections. J Rehabil
interface that relied on the Google’s accessibility tools) was Res Dev 1999; 36(4): 341-55.
[4] Kutiyanawala A, Kulyukin V, Bryce D. On Self-sufficiency of
the most time efficient for our sample of participants while verbal route directions for blind navigation with prior exposure. In:
UI2 (the touchscreen interface) was the least time efficient. Proceedings of the 25th Annual International Technology and
The qualitative user feedback showed that five out of seven Persons with Disabilities Conference (CSUN 2010), San Diego,
participants preferred UI2. We conjectured that this CA 2010.
[5] Kane S, Bigham J, Wobbrock J. Slide Rule. Making mobile touch
incongruence may be explained by a semantic dissonance screens accessible to blind people using multi-touch interaction
between our UI2 gestures based on horizontally structured techniques. In: Proceedings of 10th Conference on Computers and
hierarchies and the common understanding of the term Accessibility (ASSETS 2008), October, Halifax, Nova Scotia,
“hierarchy” which implies a vertical structure. Canada 2008; pp. 73-80.
[6] Zhao S, Dragicevic P, Chignell M, Balakrishnan R, Baudisch P.
Our semantic dissonance conjecture indicates that touch Earpod. Eyes-free menu selection using touch input and reactive
gestures should, at least to some extent, reflect the semantic audio feedback. In: Proceedings of the Computer-Human
understanding of the task prevalent in the target population. Interaction (CHI 2007). ACM Press 2007; pp. 1395-404.


[7] Vanderheiden GC. Use of audio-haptic interface techniques to [10] D. Burton. Can an Android Make Your Mobile Phone Accessible?
allow nonvisual access to touchscreen appliances (Poster). In: (2010). AFB AccessWorld, Available from: http://www.afb.org/afb
Proceedings of the Annual Meeting of the Human Factors and press/pub.asp? DocID=aw110202.
Ergonomics Society Annual 1996; 40: pp. 1266. [11] Speech Enabled Eyes-Free Android Applications. Available from:
[8] Landau S, Wells L. Merging tactile sensory input and audio data by http://code.googl e.com/p/eyes-free.
means of the Talking Tactile Tablet. Proceedings of EuroHaptics [12] Kulyukin V, Kutiyanawala A. Accessible Shopping Systems for
2003, IEEE Computer Society; 2003: pp. 414-8. Blind and Visually Impaired Individuals: Design Requirements and
[9] Li KA, Baudisch P, Hinckley K. Blindsight. Eyes-free access to the State of the Art. J Rehabili 2010; 2: 158-68.
mobile phones. in: proceedings of the computer-human interaction [13] SAS Software 9.2 (TS2M2), SAS Institute Inc., Cary, NC, USA.
(CHI 2008), ACM Press 2008; pp. 1389-98.

Received: March 21, 2011 Revised: May 10, 2011 Accepted: May 10, 2011

© Kulyukin et al.; Licensee Bentham Open.
This is an open access article licensed under the terms of the Creative Commons Attribution Non-Commercial License (http: //creativecommons.org/licenses/by-
nc/3.0/) which permits unrestricted, non-commercial use, distribution and reproduction in any medium, provided the work is properly cited.

Efficiency or Quality of Experience: A Laboratory Study of Three Eyes- Free Touchscreen Menu Browsing User Interfaces for Mobile Phones

More Related Content

Similar to Efficiency or Quality of Experience: A Laboratory Study of Three Eyes- Free Touchscreen Menu Browsing User Interfaces for Mobile Phones (20)

More from Vladimir Kulyukin (20)

Recently uploaded (20)

Efficiency or Quality of Experience: A Laboratory Study of Three Eyes- Free Touchscreen Menu Browsing User Interfaces for Mobile Phones