Content uploaded by James R. Lewis
Author content
All content in this area was uploaded by James R. Lewis on Aug 07, 2014
Content may be subject to copyright.
INTERNATIONAL JOURNAL OF SPEECH TECHNOLOGY 2, 71-79 (1997)
© 1997 Kluwer Academic Publishers. Manufactured in The Netherlands.
Design Decisions for a Voice Navigation System
MARIA MILENKOVIC
IBM Corporation, 1555 Palm Beach Lakes Blvd., W. Palm Beach, FL 33401
maro_m@vnet.ibm.com
ALAN J. HAPP
IBM Corporation, P.O. Box 12195, 3039 Cornwallis, Research Triangle Park, NC 27709
alanhapp @ vnet.ibm.com
JAMES R. LEWIS
IBM Corporation, 1555 Palm Beach Lakes Blvd., W. Palm Beach, FL 33401
jimlewis @ vnet.ibm.com
Received May 17, 1996; Accepted November 13, 1996
Abstract. Voice navigation systems allow users to interact with computer applications by voice. In this paper we
present human factors research that evaluated design alternatives for major user interface components. In the first
study we evaluated the appropriateness and discriminability of four sets of icons that were candidates for functions
in the Voice Toolbar of the navigation system. In the second study we studied two different ways of organizing and
presenting voice commands in the What Can I Say window. The key finding was that the alternative organization
seemed to improve task performance. The results provided the basis for design recommendations.
Keywords: speech recognition, voice navigation, human factors, usability, ease of use, command presentation,
icon evaluation, command organization
Introduction
Significant advances in accuracy and performance in
automatic speech recognition have resulted in several
commercial voice navigation and dictation systems. A
voice navigation system allows a user to interact with
the desktop and other computer applications by voice.
In this paper we present human factors research that
evaluated design alternatives for major user interface
components and guided some of the implementation
decisions for a voice navigation system. Together with
dictation, the voice navigation system is part of IBM's
family of speech products. This is a speaker indepen-
dent, continuous speech, large vocabulary, command
and control system. A user does not need to "train"
the system to recognize his or her voice and can use
a large number of continuous phrases to control com-
puter applications. The design reflects the need to con-
verge three lines of speech offerings that were based on
different speech technologies, developed by different
groups within IBM, and intended for different operat-
ing environments.
Several assumptions underlie the design of the voice
navigation system. First, we assumed that voice navi-
gation is part of a multimodal interface, where speech
is a valid input modality along with keyboard, mouse,
or pen. Although it attempts to capitalize on the advan-
tages of speech, the design is not for a "speech only"
interface. Voice navigation and dictation represent two
modules of functionality that need to complement each
72 Milenkovic, Happ and Lewis
other with seamless integration. We assumed two lev-
els of users: an "average" and a more "experienced"
user. The "average" user is familiar with business ap-
plications needed for a job. A more "experienced" user
is familiar with prevalent operating environments and
conventions (GUIs, desktop metaphors) and wishes to
use voice to complement or augment typical tasks.
Design Objectives
The design objectives for the voice navigation system
included the following:
• Specifying the visual elements of the interface.
• Specifying the user interaction with the system.
• Designing a voice toolbar to integrate the modular
speech functions of navigation and dictation at the
level of the interface.
• Supporting a visually consistent and unified "look
and feel" for this family of speech products across
platforms and operating environments.
The final design of the voice navigation system reflects
the influence of several factors:
• Previous expert evaluation of competitive voice nav-
igation systems. The evaluation yielded a set of de-
sirable design elements and usability criteria for a
"best of breed" navigation system.
• User interface guidelines applied to the design of
user interfaces in general and those supporting
speech as an input modality.
• Human factors research to evaluate and validate de-
sign alternatives.
Two main user interface components received particu-
lar consideration during the design process: the Voice
Toolbar and the What Can I Say Window.
The Voice Toolbar, shown in Fig. 1, consists of sev-
eral user-configurable elements:
Figure 1. An illustration of the Voice Toolbar.
• A row of buttons that provides immediate access to
speech functions, such as "What Can I Say", for a list
of valid voice commands, "Where Can I Go", for a
list of voice-accessible programs, "Begin Dictation"
to start the dictation function, and "Voice Settings",
to access voice recognition properties and settings.
• A speech volume indicator.
• A speech feedback area, where the system echoes
what was recognized or misrecognized.
The What Can I Say window displays a list of valid
voice commands the user can say at any time. The
organization of the displayed information is of utmost
importance, as users consult the What Can I Say win-
dow not only to review valid voice commands, but to
see how to complete a task using voice.
Human Factors Studies
Throughout the design and implementation process,
human factors research helped evaluate design alterna-
tives and guide decisions. In this section we will report
the findings from two studies. The first study evaluated
four sets of icons which were candidates for functions
in the Voice Toolbar (Ehrlich et al., 1995). The second
study (Milenkovic et al., 1995), evaluated the usability
of an alternative organization of the What Can I Say
window.
Study One: Evaluating the Appropriateness of Four
Sets of Icons for a Speech Recognition Toolbar
In the first study, we assessed the appropriateness and
discriminability of four sets of icons. The Microphone
Off icon indicates that the microphone is off and the
voice recognition system can not respond until the mi-
crophone is on. The Microphone Asleep icon indicates
that the voice recognition system does not respond un-
til the user says "wake up". The "Where Can I Go"
icon opens a window with a list of programs and ap-
plications the user can access by voice. The Dictation
mode icon is a visual indicator that the application is
in dictation mode rather than navigation or command
and control mode.
The IBM Icon Reference Book (1991) provided
guidelines for testing the appropriateness of icons. The
testing method used a paired comparison procedure to
assess both appropriateness and discriminability of the
icons. The dependent measure was a score based on
Design Decisions 73
the subject's preference for one or the other icons or
both (i.e., there was no difference in preference).
Method
Subjects.
Fourteen subjects participated in the study.
One group consisted of seven people hired from a lo-
cal agency, described as experienced application users,
but not experienced with speech products. The second
group consisted of IBM employees who were current
users of voice navigation and dictation products.
Icons.
Table 1 shows the four sets of icons that
the participants evaluated. The
Microphone Off
and
Microphone Asleep
icons differ in the orientation of
the microphone component (diagonal or horizontal)
and the background color used (either gray or yellow).
Two of the icons with the horizontal microphone were
yellow and one had the letters "Zzz" in black shown
above the microphone (icon #3 and icon #4). The
Dic-
tation Mode
icons all used the voice bubble and colored
characters: red in icon #2, blue in icon #3. In the
Where
Can I Go
icon set, the arrows were red. In icon #2 of the
set, the question mark and the window title were teal 1 .
An Authorware
TM
program presented the icons. The
program randomly selected each set without replace-
ment and randomly presented pairs of icons without
replacement. Each icon pair appeared in a separately
labeled display. For a set of 4 icons, e.g., the
Micro-
phone Off
set, there were 6 different displays. The
top of each display showed instructions and the con-
cept that the participant was to evaluate. The display
showed the pair of icons in a numbered list.
Procedure.
Participants completed a short demo-
graphic questionnaire that collected information on
computer use and, specifically, use of speech recog-
nition products. When ready, a participant pressed the
space bar to start the paired comparison test. For each
comparison, a participant stated whether the icon la-
beled "1" or "2" better represented the concept being
evaluated. A participant could respond "both" or "nei-
ther" if there was no preference for either of the icons
(i.e., they were equivalent). The experimenter recorded
all responses from a participant on a response sheet.
Participants worked at their own pace. They completed
the comparisons for a set of icons before starting a new
set; after each set, the experimenter asked the partici-
pant for suggestions of other alternatives.
Dependent Measure.
The test methodology pre-
scribes the following algorithm for scoring a partic-
ipant's responses: "for each icon pair displayed, the
most appropriate icon received a score of "2" and the
least appropriate icon received a score of "- 1". When
neither icon was preferred over the other, each icon in
the pair received a score of "1". (IBM Icon Reference
Book, 1991, pp. 15-21). For the icon sets with four
icons, the range of possible scores was 9, from "-3"
to a maximum of "+6". For the Dictation Mode icon
set, the minimum score was "-2" while the maximum
score was "÷4". For the "Where Can I Go" icon set,
the maximum preference score was "+10", while the
lowest possible preference score was "-5".
Results
Microphone Off Icon Set.
Table 2 shows the mean
and standard deviation of the preference scores of
the
Microphone Off
icon set for both groups of
participants. A repeated measures analysis of vari-
ance (ANOVA) with experience and icon as factors
showed a marginally significant main effect of icon
(F(3, 36) = 2.28, p = 0.10) and experience by icon:
(F(3, 36)=2.36, p = 0.09). Overall, participants
Table 1.
Four icon sets used in the experiment.
Microphone Off and Microphone Asleep
1 2 3 4
Dictation
Q...O
1 2 3
Where Can I Go
m
1
m
2 3
m
4
m
5
74 Milenkovic, Happ and Lewis
Table 2. Preference scores for the icons evaluated for the concepts Microphone Off and Microphone Asleep.
~ons
A
l
B
J
C
D
Microphone Off Micxopl~
Experience
Level
HI
lo
HI
Io
HI
1o
:HI
1o
Mean
1.29
-0.86
4.43
2.14
-0.14
3.00
1.71
3.14
Star Dev.
3.40
3.39
2.22
2.16
3.08
3.46
1.98
2.67
Mean
-1.14
-1.00
0.86
0.57
4.29
5.86
2.57
2.29
le Asleep
Std Dev:
3.48
2.00
1,21
0.53
3.40
0.38
1.90
1.50
preferred Icon B for Microphone Off. The interaction
was marginally significant because the pattern of pref-
erence differed for experienced and less experienced
users. Experienced users strongly preferred Icon B,
but less experienced users expressed satisfaction with
Icons B, C, and D.
Microphone Asleep Icon Set. Table 2 also shows the
mean and standard deviation of the preference scores
of the Microphone Asleep icon set for both groups of
participants. A repeated measures analysis of vari-
ance (ANOVA) with experience and icon as factors
showed a significant main effect only for the icon factor
(F(3, 36) = 16.57, p < 0.001). Overall, participants
preferred Icon C to represent Microphone Asleep.
Dictation Mode Icon Set. Table 3 shows the mean and
standard deviation of the preference scores of the Dic-
tation Mode icon set for both groups of participants. A
Table 3. Preference scores for the icons evaluated for the concept
dictation mode. Dictation
F.xperlence
Icons Level Mean Std Dev.
A 0...0
B®
cAO
HI - 1.71 0.76
1o 1.86 1.57
]t1 2.57 1.51
lo 1.29
2.21
HI 2.86 1.35
1o 0.29 2.14
repeated measures analysis of variance (ANOVA) with
experience and icon as factors showed a significant ef-
fect for the icon factor, and a significant interaction
(F(2, 24) = 3.55, p = 0.04, and F(2, 24) = 9.58, p <
0.001, respectively). Overall, participants preferred
Icon B. An examination of the mean preference scores
shows that the significant interaction occurred because
the icons most preferred by the experts (B and C), were
least preferred by the novices (who most preferred A).
Where Can I Go Icon Set. Table 4 shows the mean
and standard deviation of the preference scores of
the Where CanlGo icon set for both groups of par-
ticipants. A repeated measures analysis of variance
Table 4. Preference scores for the icons evaluated for the concept
Where Can I Go. Where Can !
Go
Icons
Aa
BI
DII
El
a1
F i!
Experience
Level Mean Std Dev.
HI 1.86 1.86
Io 3.71 2.21
HI 2.00 4.24
1o 4.86 1.68
HI 2.57 1.99
Io 1.71 3.30
HI 9.86 0.38
Io 8.14 2.16
HI 3.00 1.73
1o 2.57 4.24
HI 3.86 2.34
1o 1.71 2.87
Design Decisions 75
(ANOVA) with experience and icon as factors showed
a significant main effect only for the icon factor
(F(5, 60) = 12.34, p < 0.001). Participants strongly
preferred Icon D.
Discussion. One goal of this study was to facilitate the
adoption of an appropriate icon for selected functions
of a speech product. Four sets of icons were evaluated:
Microphone Off, Microphone Asleep, Dictation, and
Where Can I Go. The decision to recommend one of the
icons over the other rests not only on the results of the
statistical analyses but also on design tradeoffs that af-
fect the look and feel of the entire interface. One trade-
off was that the responses of the experienced speech
recognition users were given more importance than
those of the less experienced users. There were two
reasons for this. First, the experienced users have a
fuller model of speech recognition tasks so selecting
icons that they prefer might foster that model in novice
users, resulting in faster learning. Second, novice users
are more likely to have the help "bubbles" turned on so
that the system presents the function of the icon in text
to the user. As novices learn the system, they may turn
that help off, and at that point, they should have the most
appropriate icon available. Another tradeoff is that any
individual icon must fit with the look and feel of the
interface. This means that even with a strong prefer-
ence by users for an icon, the design team may make
changes in the look of the icon to be sure it is consistent
with the other icons in the application or system. Re-
search has shown that the concept represented is more
important than details such as color (Miller et al., 1992)
or prototypical features (Byrne, 1991). The results of
the study, along with other tradeoffs, suggested the fol-
lowing icons be used in the Voice Tool bar:
Microphone Off Microphone Asleep
From an end user and system perspective, the Mi-
crophone Off and the Microphone Asleep icons must
be considered together. The Microphone Off icon has
a gray background while the Microphone Asleep icon
has a yellow background. The novice users liked the
icons with the yellow background better than the icons
with the gray background regardless of the concept
being evaluated. However, it would be better to take
advantage of a difference in color of the icons. The
experienced speech recognition users clearly preferred
the recommended icons which have different back-
grounds. The result of the ANOVA for the Microphone
Off concept suggested trends for significant differences
among the icons and differences depending on the type
of user (the significant interaction). The recommended
Microphone Off icon reflects the decision to use the
icon that was slightly less preferred by the novice users
since the system could provide help tags (text that pops
up with the icon's label) and the recommended icon
would help the novices learn the preferences of the ex-
perienced users.
Dictation
AQ
Interestingly, this icon was the most preferred by the
experienced speech users but the least preferred by the
novices. Two reasons support the recommendation to
use this icon. First, it is most similar to the graphic
on the button that opens the IBM Voice Type dictation
window. Users should benefit from the consistency.
Secondly, by using the icon preferred by the experi-
enced users we might help teach the novice users to
see the system more like an experienced user. Help
tags, on initially by default, teach users what the func-
tions are while they are learning the system.
Where Can I Go
i
Both groups of users endorsed the concept of a compass
crosshair icon for the Where Can I Go button. (None
of the participants mentioned it, but the directions for
E and W are, in fact, mislabeled. The actual icon will
have the compass directions properly labeled !)
In summary, the icon test methodology worked well
for the evaluation of these icons. For small sets of icons
this procedure can be done quickly and without fatigu-
ing the participant. However, when the number (n) of
icons increases, the sheer number of paired compar-
isons, which increases as (n * (n - 1))/2, makes the
test a chore for the experimenter and the participant.
Often, the real question is whether users find the set
of all icons in a product appropriate and discriminable.
When this is the case, methods used by Lewis (1993a),
Fullerton and Happ (1993), or Lin (1992) should be
considered.
76 Milenkovic, Happ and Lewis
Second Study: Human Factors Evaluation
of an Alternative "What Can I Say"
Window Organization
One current use for continuous speech systems is to en-
able people to issue verbal commands to control com-
puter applications. This allows the use of the natural
communication mode of speech. However, systems
that lack perfect grammatical parsing and interpreta-
tion of the speech input (in other words, all modern
commercial computer speech systems), require users to
know what phrases the speech-enabled application can
accept as valid commands. Thus, most command-and-
control continuous speech systems provide, on com-
mand, a list of the phrases that the system can currently
understand--a What Can I Say window. When a user
isn't sure what to say next, he or she can browse the
list of commands in this window.
The objective of the study was to evaluate the us-
ability of an alternative organization of the
What Can
I Say window for the IBM
TM
Continuous Speech
System
TM
(ICSSTM), as displayed in the Human-
Center, the multimodal user interface prototype for
IBM's Personal Computer Power Series
TM.
Specifi-
cally, we sought to determine whether the alternative
organization improves a user's ability to find a desired
command. In both organizations, voice commands are
organized and displayed under headings that indicate
logical groupings. A small button with a "+" sign
in front of the heading indicates that the group can
be expanded or reduced to the principal heading by
clicking the plus sign or by saying the name of the
heading. For hiding the contents of the group, the
user can click the button now displaying the minus
sign, or say the name of the heading again. Previ-
ous research (Chimera and Shneiderman, 1994) indi-
cates that a comparable "expand/contract" scheme or
a multipane interface scheme, vs. a fully expanded
interface, produces significantly faster times for in-
formation retrieval tasks. Relative to the initially de-
signed organization, the alternative organization has
fewer high-level categories, with the categories orga-
nized to focus more on likely user tasks. Figure 2 illus-
trates the appearance of the originally designed What
Can
I Say window, and Fig. 3 shows the alternative
organization.
Lv, me
Figure 2.
window.
Action
Humam
Center
-~oilap#e Human Centre
i~Expand Human Cente~ Main Window
~Expand Human Center Shodcut#
GBExpaxKI Pro~am Camtlol Menu
-3Collapse Global
GCollapse Speech Centre
Kincaid
Go-To-Sleep
Microphone-O ff
Wlkme-Cam4-Go
What-C~-I-Say
Learn CommarKl#
Speech Settie~#
BExpa¢¢l Proglaem~
BExpamd What-Can-I-Saje Commands
(~lExpand Getting Help
(~Expand Kayboa~d
(~Expand Clipboard
I~Expand Dialog Conlkots
g]Expand Tmd Editing
~Ezq~md Text Selection
~J I!:I i ]~ i ill III I I I IK~[ lJ i ,1- [I}.'J[i~l I I[i,I ill
m
,,, ,,,,,,,,,, ,
An illustration of the initially designed
What Can
I Say
Action
-~Collapse ALWAYS ACTIVE
G
Chmlie
Go-To-Sleep
Microphone-Off
Whine-Cam-I-Go
What-Cam-I-Say
Fine Tune Recognitiem
L:]Collap~m WHAT -CAN-I -SAY ACT I0 N S
Switch-To What-Can-I-Sa~
Close
What-Cam-I-Sa~,
Add Program Macro
Add Global Command
T rain Word#
Redefine Program Macro
Modify Program Properties
Delete Command
Delete Program Properties
Learn
Command=
Pri~ What-Can-I-Say
Command=
mcee.pamd> What-Cared-Say
Beench
B]lExpand PROGRAM
ACTIONS
I~Expamd EDITING COMMANDS
~EHIpa~I CURSOR
MOVEMENT
mExpamd HELP COMMANDS
¥
Figure 3. An illustration of the alternative
What Can
I Say win-
dow.
Design Decisions 77
Method
Subjects.
Eight IBM employees (4 male and 4 fe-
male) participated in the study. All participants were
native American English speakers and familiar with
graphical user interfaces and IBM's voice recognition
products.
Equipment.
The Human Center ran on an IBM Per-
sonal Computer Power Series Model 6015 (with the
601/66 MHz processor) using Windows NT TM as the
operating system.
Procedure.
There were two types of tasks used in this
study 2. For the Explicit voice command tasks, the par-
ticipants received a card which contained a voice com-
mand for which they were to search. In the Scenario
tasks, participants read a short scenario describing a
navigation task. They then searched for the appropriate
voice command needed to complete the task. Each par-
ticipant completed five of each type of task (Explicit,
Scenario) for each type of What Can I Say organi-
zation (Current, Alternative). A Greco-Latin design
(Lewis, 1993b) was used to counterbalance the order
of presentation of task sets (a, b), type of task, and type
of What Can I Say window. Table 5 shows the explicit
phrases and scenarios used in this study.
To begin the experimental session, the experimenter
explained the procedure to the participant. For each
task: The experimenter demonstrated how to com-
plete the task and handed the participant a card de-
scribing the task. The participant then searched for
the appropriate voice command in the What Can I
Say window. The experimenter recorded the time re-
quired to complete the search (Task Time), the number
of times the participant clicked the mouse to expand or
contract a high-level heading (Heading Clicks), and the
number of times the participant clicked the mouse to
move the information in the window (Scrolling Clicks).
The participants completed the tasks in sequence for
each What Can I Say window organization as assigned
by the Greco-Latin experimental design. After com-
pleting the tasks for one What Can I Say organization,
there was a short break as the experimenter loaded the
other What Can I Say organization. After participants
finished all the tasks, they completed a short question-
naire which allowed them to comment on their experi-
ence.
Table 5.
Descriptions of the experimental tasks.
Number Explicit voice commands Scenario
la
Move Window Right 20
2a Always on Top
3a Back Tab 20
4a Previous Program
5a Top of Document
lb Enter-Key
2b Next Field 20
3b Select Complete Document
4b Speech Center Help
5b Page-Up 20
You want to move to the end of the line that you're typing on. Find the
command for this.
Someone has come into your office and you want to switch offyour
microphone so the navigator won't pick up your conversation.
Find the command for this.
You want to activate the Escape key by voice. Find the command you
need to use.
You want to activate the Tab key 10 times. Find the command that
allows you to do this.
You are tired of looking at the What Can I Say window. Find the
command that allows you to close it.
You are typing out a filename to save it and you want to find the
command that changes from insert to typeover mode.
You are reading through a paper you've written in Write
TM.
You want
to check the references at the end. Find the command that moves you
to the very end.
You want to delete the line you're typing, but first you need to select it.
Find the command that allows you to select it.
You are creating a large drawing document and you want to move down
a page. Find the command that lets you move down one screen.
You are constantly opening the same two applications together and you
want to create a voice command that will do this for you. Find the
command that allows you to do this.
78 Milenkovic, Happ and Lewis
Results
Search Time. Figure 4 shows the results for the time
required to search for the target phrase as a function
of type of What Can I Say organization (Current,
Alternative) and Stimulus Condition (Explicit, Sce-
nario). For the Explicit task, participants found the
target phrase, on average, 36% faster with the alterna-
tive organization. For the Scenario task, the reduction
in search time was an average of 23%. Although these
effects appear relatively large, they were not statisti-
cally significant. The effects of stimulus condition and
the organization by stimulus interaction were also non-
significant. The reasons for the variability in search
times are the small sample size and a strong, signifi-
cant learning effect across trials, especially from trial
1 to 2. Part of what we have learned from conducting
this study is that, given a small sample, it would be
useful to provide more practice than we did to stabilize
performance before starting the experiment proper.
Mouse Clicks. The number of mouse clicks on head-
ings within the What Can I Say window is another
measure of how easy or difficult it is to find a specific
voice command within the What Can I Say window.
When clicking on a heading, the heading expands and
reveals the valid voice commands grouped underneath.
The assumption is that a greater number of mouse clicks
40.0
35.0
30.0
25.0
A
2oo
F-
15.0
10.0
5.0
0.0 Explicit
Scenario
Type of Task
Figure 4. Mean search times as a function of the What Can I
Say
organization and type of task.
indicates a greater difficulty in locating a command be-
cause the participant has expanded several headings to
find a particular command. The number of scrolling
clicks was a similar measure. It reflects how many
times a participant clicked to scroll the whole What
Can
I Say window while trying to locate a command.
The key finding is that the total number of mouse clicks
was equal between the two organizations Thus, it is not
reasonable to attribute differences in search times to
differences in mouse click requirements between the
two organizations.
Commands That Were Difficult to Locate. To identify
which commands were most difficult to locate in the
alternative organization, we performed item analyses
to identify:
• Commands with higher than average search time
• Commands with higher than average heading clicks
• Commands with higher than average scrolling clicks
These criteria indicated the potential for improve-
ment in the location of several commands. For exam-
ple, five commands (Top of Document, Microphone
Off, Escape Key, Speech Center Help, Select Line)
had measurements below average for all three crite-
ria. Thus, these commands would be prime candidates
for relocation in a subsequent redesign.
Discussion. The main finding of this study was that the
alternative organization seemed to improve task per-
formance, with an average reduction in search time of
36% for the first type of task and an average reduction
of 23% for the second type of task. Thus, the results
suggest that the alternative organization is better than
the initially designed organization. In addition to sup-
porting the adoption of the alternative What Can I Say
window organization, the data point to specific ways
to further improve the alternative organization by re-
locating certain commands (identified in the Results
section). Failure to reach statistical significance for
differences in search time is probably due to the small
sample size and high performance variability. Because
the apparent effect size was large, we did not replicate
the study to increase the sample size in an attempt to
achieve statistical significance.
Conclusions
Traditional human factors methods and processes were
applied while iterating through the design of a voice
Design Decisions 79
navigation system and two of its major user interface
components, the
Voice Toolbar
and the
What Can
I
Say window. The evaluation of four sets of icons for
discriminability and overall appropriateness provided
the basis for recommending one icon in each set for
the speech functions under consideration (Microphone
Off, Microphone Asleep, Dictation mode, and Where
Can I Go). The alternative What Can I Say window
organization, with fewer higher level categories and
categories organized to focus more on likely user tasks,
seemed to improve task performance. We plan to pur-
sue further exploration of alternative ways to structure
and present valid voice commands to the user. Over-
all, subsequent usability enhancements to the organi-
zation of the
What Can
I Say window, as well as to
the logical flow and sequencing of the more difficult
tasks (creating a voice macro), will be incorporated
in subsequent design iterations of the voice navigation
system.
Acknowledgments
We wish to express our appreciation to Jennifer Ehrlich
and Lise Schneider who helped to design the experi-
ment and collected the data for evaluating the appropri-
ateness of four sets of icons. We also wish to express
our appreciation to Ron Van Buskirk, who participated
in the design of the alternative What Can I Say organi-
zation, helped to design the experiment, and collected
the data for the second study.
Trademarks
IBM, Personal Computer Power Series, IBM Contin-
uous Speech System, and ICSS are trademarks of the
International Business Machines Corporation. Author-
ware is a trademark of Macromedia Inc. Windows NT
and Write are trademarks of Microsoft Corporation.
Notes
1. The icons were designed for the US product. National Language
Support issues should be taken into consideration for international
products.
2. The tasks were not necessarily directly comparable, but the design
of the experiment and subsequent analyses did not require them to
be comparable. Instead, they were designed to cover a reasonable
range of tasks.
References
Byme, M.D. (1991). The misunderstood picture: A study of icon
recognition. In Proceedings of Human Factors in Computing Sys-
ten~ (CHI '91). New York: Association for Computing Machin-
ery, p. 493.
Chimera, R. and Shneiderman, B. (1994). An exploratory evalu-
ation of three interfaces for browsing large hierarchical tables
of contents. ACM Transactions of Information Systems (USA),
12(4):383-406.
Ehrich, J., Happ, A.J., and Schneider, L. (1995). Evaluating the
appropriateness of four sets of icons for a speech recognition tool
(Tech. Report TR54.897). Boca Raton, FL: IBM.
Fullerton, S. and Happ, A.J. (1993). A user-oriented test of icons
in an educational software product. In G. Salvendy and M.J.
Smith (Eds.), Proceedings of the Fifth International Conference
on Human-Computer Interaction (HCI International '93). Ams-
terdam: Elsevier, pp. 44-49.
IBM Corporation (1991). IBM Icon Reference Book. (SC34-4348-
00). Cary, NC: Author.
Lewis, J.R. (1993a). An icon usability evaluation procedure
with application to personal communicator icons (Tech. Report
TR54.792). Boca Raton, FL: IBM.
Lewis, J.R. (1993b). Pairs of Latin squares that produce diagram-
balanced Greco-Latin designs: A BASIC program. Behavior Re-
search Methods, Instruments and Computers, 25:414-415.
Lin, R. (1992). An application of the semantic differential to icon
design. In Proceedings of the Human Factors Society 36th Annual
Meeting. Santa Monica, CA: Human Factors and Ergonomics So-
ciety, pp. 336-340.
Milenkovic, M., Lewis, J.R., and Happ, A.J. (1995). Human factors
evaluation of an alternative "What Can 1 Say" window organization
for ICSS (Tech. Report 54.907). Boca Raton, FL: IBM.
Miller, M.E., LaLomia, M.J., and Happ, A.J. (1992). Icon search pro-
cesses: Effect of icon color and category cue on eye movements,
identification time, and accuracy (Tech. Report TR54.659). Boca
Raton, FL: IBM.
