EFFICIENT NAVIGATION ON THE WORLD WIDE WEB FOR THE
PHYSICALLY DISABLED
Leo Spalteholz, Kin Fun Li
Department of Electrical and Computer Engineering, University of Victoria, Victoria, BC, Canada
Nigel Livingston
University of Victoria Assistive Technology Team, University of Victoria, Victoria, BC, Canada
Keywords:
Web accessibility, single switch input, web navigation, physical disability.
Abstract:
One of the major obstacles with current web access solutions for those with physical disabilities is the efficient
selection of links and other web page elements. This is especially so for users of single switches. Current
solutions simply build a list of the selectable web page elements and use a linear scan to iterate through them,
selecting the highlighted link when the user activates an input device. We propose a new method based on
incremental searching of the link text to select elements. This approach, paired with well-established switch
typing solutions, allows a single switch user to select any element on a web page by typing very few (most
often only one) letters.
1 INTRODUCTION AND
MOTIVATION
For many people with limited dexterity or severe
physical disabilities, interaction with a computer re-
quires a specialized input device. The choice of which
input device to use is highly dependent on the users’
level of physical control. For example, for those at
the advanced stage of Amyotrophic Lateral Sclerosis
(ALS), eye muscles are often among the last to de-
teriorate, making the use of an eye tracking system
suitable to control the computer mouse pointer. For
other disabilities, such as muscular dystrophy, one or
more hardware switches, activated by any body part
capable of consistent, voluntary movement, may be
used along with specialized software to control the
computer. For the severely disabled, a single switch –
although it is a low bandwidth input is a widely used
control device due to its simplicity and economy. A
switch may take the form of a physical button, a relay
activated by a puff of air through a tube, or a switch
activated by the electrical signals produced by muscle
cells as in electromyography (Prinz et al., 2006). The
problem of how to produce text with a single input
has been thoroughly researched, with the first typing
devices for single switch users being proposed almost
30 years ago (Rowell et al., 1978).
While typing text is still a key component of com-
puter access, efficient access to the information on
the world wide web is becoming central to the use
of computers. Especially for those with mobility im-
pairments, the world wide web represents an excit-
ing opportunity to move beyond the limitations they
may face when interacting with people in the out-
side world. To that end, it is critical that the world
wide web be accessible for those with physical im-
pairments. Although standards for accessible web
pages are an important component of the overall ef-
fort towards universal accessibility, for example the
W3C Web Accessibility Initiative (w3.org, 2006), the
reality is that the majority of web pages are not devel-
oped with accessibility in mind and this statistic is not
improving (Hackett et al., 2004). The current state of
accessibility support on the world wide web requires
tools that can handle any web site, regardless of its
structure or adherence to standards.
From the authors’ experience with users of low
bandwidth input devices as well as with their aides
(University of Victoria Assistive Technology Team,
UVATT.org), it is apparent that although there are
many people that use their devices for computer con-
trol or text input, the web has remained largely inac-
cessible to them. Judging from requests received for
assistance in accessing the web by users with a wide
321
Spalteholz L., Fun Li K. and Livingston N. (2007).
EFFICIENT NAVIGATION ON THE WORLDWIDEWEB FOR THE PHYSICALLY DISABLED.
In Proceedings of the Third International Conference on Web Information Systems and Technologies - Web Interfaces and Applications, pages 321-327
DOI: 10.5220/0001287703210327
Copyright
c
SciTePress
range of disabilities, we can conclude that an effec-
tive web access method would be useful to many. As
there is no lack of either potential users or demand for
a usable web access system, the fact that it is so rarely
seen in practice may be partially due to the inadequate
efficiency of currently available solutions. In our ex-
perience testing various accessible web browsing sys-
tems, the large number of links and the complex lay-
out in modern web pages makes navigation unaccept-
ably slow and cumbersome.
To address this issue, we propose a new technique
for selecting web page elements for bandwidth lim-
ited input devices. This method will be a key com-
ponent in a customized browser designed to be used
with alternative, low bandwidth input devices. Sec-
tion 2 discusses some of the previous work in this
area, with particular emphasis on the linear scanning
selection method (Section 2.1) for selecting elements
as employed by many previous approaches. The ma-
jor components of our proposed system are detailed
in subsections of Section 3, including specifics of the
implementation.
2 BACKGROUND SURVEY
One of the cornerstones of computer use has always
been, and continues to be entering text into the com-
puter. This has prompted a wealth of work on cre-
ating efficient typing interfaces that can be operated
with a single input signal such as that generated by
an alternative input device. A common approach to
this typing interface has remained relatively constant
from its inception, with letters being selected from a
grid by automatic scanning and typed upon activation
of the input device. The most efficient implemen-
tations use a divide and conquer approach by divid-
ing the available characters into groups and narrowing
down the selection on each switch activation. Com-
monly found in commercial switch scanning prod-
ucts (madentec.com, 2006) is the row-column scan-
ning approach where letters are displayed in a two di-
mensional matrix and scanned through first by row,
and then, upon switch activation, by column until the
desired letter is selected (Steriadis and Constantinou,
2003).
Conversely, the primary method for navigating on
the world wide web has traditionally been by “point
and click”. This method is not easily adaptable to sin-
gle input devices that are not capable of accurate two
dimensional input. One approach to solving this prob-
lem is to use several on-screen buttons to move the
mouse pointer on the screen. For single switch users,
this is usually done by presenting a button for each
direction of mouse movement (x and y, as well as di-
agonal) and then scanning through the buttons. While
this method is a valuable fallback for applications that
are not accessible via the keyboard, it is not an ef-
ficient interface for continuous use, and requires pre-
cise timing control in switch activations from the user.
The inefficiency of these pointer control schemes has
prompted research into other ways to navigate on the
world wide web.
There are two notable attempts at creating a web
browser specifically accessible to single switch users,
the AVANTI browser (Stephanidis et al., 1997) and
the MultiWeb browser (Owens and Keller, 2000).
Both of these approaches use a linear scanning ap-
proach to link selection, with some differences in how
the link scanning is presented to the user. In addition,
(Mankoff et al., 2002) developed a web browser ac-
cessible to users capable of operating four separate
switches using a similar linear scanning approach,
with focus being advanced by switch activation in-
stead of automatically on a timeout. The linear scan-
ning technique is discussed more thoroughly in Sec-
tion 2.1. Hanson et al. with their accessibilityWorks
project, while focusing more on adapting the appear-
ance of web pages, also consider the issue of con-
trol by devices with zero or one dimensional input
(Hanson et al., 2005). Similar to other work in the
area, they propose to extract the list of links and place
them in an external list such that they can be selected
by a user incapable of two dimensional cursor con-
trol. The Gnome On Screen Keyboard (gok.ca, 2006)
implements a similar link extraction scheme for web
browsers that can expose web page structure to an ac-
cessibility framework (Zhao et al., 2005).
In more mainstream web browsers, two alternate
approaches to linearly iterating through links on a
web page have emerged: incremental search and au-
tomatic access keys. Mozilla Firefox implements a
feature called “find as you type”, which incremen-
tally searches a web page and can highlight match-
ing links as one types a search query (mozilla.org,
2006a). If only one link matches a search, it can be
navigated to by pressing the “enter” button on the key-
board. This allows the user to select text and image
links, but does not allow focusing of text entry boxes,
clicking of form buttons, or selecting other form input
fields. Another approach is automatic access keys, as
implemented in the Konqueror (konqueror.org, 2006)
web browser on the Linux platform. The automatic
access keys approach assigns a unique alphanumeric
character to each selectable element on a web page,
and shows these characters with a tooltip over each
element. Typing an assigned letter will select the cor-
responding element. Currently, the system only as-
WEBIST 2007 - International Conference on Web Information Systems and Technologies
322
signs the 36 alphanumeric characters, meaning that
any web pages with more than 36 elements will not
be accessible.
2.1 Linear Scanning Selection
The most common approach to web page element se-
lection for switch specific browsers is to construct a
list of selectable elements in the current web page,
and then linearly scan through this list. There are two
general approaches to scanning through selectable el-
ements: in-place and external scanning. In-place
scanning, as used by the MultiWeb browser, as well
as all major mainstream browsers, scans through the
list of selectable elements on the page by highlighting
them within the original web page layout. The ex-
ternal scanning approach works by extracting all the
selectable elements in the page and displaying them
in a list beside the web page. This list is then scanned
through top to bottom, and a link can be navigated
to by activating the switch when the target is high-
lighted.
While the linear scanning approach is a logical
extension of the scanning used in switch typing in-
terfaces, it has some key disadvantages that limit its
efficacy for navigating the web. In-place scanning
has the advantage of lending context to links with
non-descriptive names. Although it is considered bad
practice in the field of web development, many hy-
pertext links are given names such as “link” or “click
here” which are only meaningful in the context of the
surrounding text. If these types of links are extracted
from the web page and scanned through in an exter-
nal list, it becomes very difficult to determine which
link is the desired target. Another disadvantage of the
in-place scanning technique is that the scanning order
can be unpredictable. Since the user must quickly ac-
tivate the switch when his/her element is selected, it
is crucial that the scanning order be predictable. Al-
though scanning elements in a top to bottom order
is logical for simple web pages, sites with a compli-
cated layout can make element scanning very difficult
to follow.
The primary disadvantage of both linear scanning
approaches is their lack of scalability to web pages
with many selectable elements. It is very common for
web pages, especially those of news networks, search
engine results pages, or site maps to contain many
tens or even hundreds of links. Linearly scanning
through large numbers of elements to find a desired
target is a very tedious process. The external scanning
approach gives more flexibility in grouping the list of
links and selecting them via a divide and conquer ap-
proach to improve performance, but is still difficult to
present to the user in an intuitive way for large num-
bers of links. The error cost of the linear scanning
approach is also high, as a missed target would, in the
best case, involve having to wait for the scan to loop
around, and in the worst case result in a click on an
incorrect link, necessitating multiple steps to return
to the previous page.
3 PROPOSED SYSTEM
OVERVIEW
Our proposed system is designed as an extension to
the open source Mozilla Firefox web browser (Sec-
tion 3.3), adding a more efficient text based naviga-
tion method for low bandwidth input. This system
will facilitate the activation of selectable elements on
a web page, such as hypertext links, text entry boxes,
form buttons, or image maps. After a web page is
loaded, the label builder (Section 3.2) constructs a
textual label for each selectable element on the page.
To select an element, the user employs a text en-
try interface (Section 4) to type the starting letters
of his/her desired element’s label. After each letter,
the selectable elements on the page are searched and
highlighted according to the algorithm described in
Section 3.1. When only a single element matches the
entered query, the user is prompted to activate (nav-
igate to) the selected element. Figure 1 depicts an
overview of the major components of the system.
3.1 Incremental Search Selection
To address the usability issues found in the linear
scanning approach, we propose an alternate approach
to selecting elements on a web page based on text en-
try instead of direct selection. Using this approach,
we can leverage the research done on efficient text en-
try using alternative input devicessuch as eye tracking
systems or single switch devices.
The incremental search selection method is based
upon selecting a web page element by typing a sub-
section of its user-visible label. For hypertext links or
form buttons, the identifying label is simply the text
of the link or button. For text input boxes and other
selectable elements with no meaningful identifier, the
system will generate a unique label to present to the
user. Our approach to automatic label generation is
discussed further in Section 3.2.
When a new page has completed loading, the sys-
tem extracts all selectable elements from the page and
builds a list of labels from them. This list comprises
the search space for the highlighter module. When the
EFFICIENT NAVIGATION ON THE WORLD WIDE WEB FOR THE PHYSICALLY DISABLED
323
Figure 1: System Overview.
link function is selected, the system waits for text in-
put from an external on-screen keyboard application
designed for alternative input devices. To select an el-
ement on the page, the user starts typing the label of
that element. When the first letter is typed, the system
performs a prefix search of all the elements and visu-
ally highlights any matches. Every subsequent letter
will narrow this search until only one element is high-
lighted. The system will then prompt the user if that
selected element should be activated. In the case of a
hypertext link, an activation entails navigating to the
linked page, while a text box would be focused, and
a button would be pressed. The goal of the highlight-
ing algorithm is to be as intuitive as possible for the
user while allowing fine grained specification of the
desired link. Figure 2 details our algorithm used to
highlight elements on the page.
In Figure 2, the step labeled “Widen Search Field”
requires some additional explanation. When the
user initially activates the link selection function, the
search field is comprised of the visible selectable el-
ements on the current screen. Thus, if a web page is
longer vertically than the size of the screen, the al-
gorithm will only match elements that are currently
visible to the user. If no elements start with the let-
ter(s) typed by the user, the search field expands to en-
compass all the selectable elements on the web page.
If there are no matching elements at this level, the
search criterion is relaxed to consider any substring
Figure 2: Search Selection Highlighting Algorithm.
of all the elements on the current page. In the vast
majority of cases, the user will be attempting to se-
lect an element that is currently visible to them. This
search fallback method maximizes the efficiency for
the common case of a user wishing to select a visible
element, while still allowing a user to select an off-
screen target if he/she has existing knowledge of its
label.
The selective case sensitivity is included to de-
crease the average number of necessary letters to
uniquely identify any given element while not impos-
ing additional constraints on the user. When using an
on-screen keyboard, much like with a physical key-
board, producing a capital letter requires pressing an
extra key. Thus, if a user types a capital letter, it is as-
sumed that it is not accidental. Performing a case sen-
sitive search when the user has typed a capital letter
opens up the set of upper case letters to be populated
with labels. For example, if a page contains multiple
WEBIST 2007 - International Conference on Web Information Systems and Technologies
324
links starting with a lowercase ’b’, and the desired el-
ement label starts with a capital ’B’, then the desired
element can be uniquely selected by typing a capital
B, instead of possibly having to type multiple letters
if the system was not case sensitive. Typing a lower
case letter matches both upper and lower case labels
since the system should not reduce efficiency in cases
where there is only a single label starting with that
letter (in upper case).
3.2 Element Labeling
There are several factors to consider for the element
label generation algorithm. The primary goal is to
minimize the number of characters necessary to se-
lect any given element on the web page. A secondary
goal is to present link text intuitively and make as few
modifications to the visual appearance and layout of
the web page as possible. Initially, the system ex-
tracts the text of every link and every button on the
page and constructs a list based on these labels. For
elements without a user-visible label, the system con-
structs a list of the alphanumeric characters that are
not already used as starting characters in the set of el-
ement labels. For each unlabeled element, a label is
created consisting of an unused alphanumeric charac-
ter prefixed to the type of the element. For example,
a text edit box might receive the label “bTextbox” or
“4Textbox” while an image link may receive the la-
bel of “tImage”. It should be noted that these kind
of labels are only constructed if no meaningful infor-
mation about the element can be determined which
could be used instead. For example, some form input
controls (check boxes, text entry boxes, radio buttons,
etc.) are referenced by an external “label” HTML tag
which provides a textual description of its purpose,
and image links may have an alternate textual label
specified.
The tradeoff to consider when constructing labels
for textual links is whether to provide as much cover-
age of the alphanumeric character set as possible ver-
sus leaving the labels as they appear on the page. Ag-
gressive relabeling ensures maximum efficiency but
necessitates more layout modifications and may re-
duce the intuitiveness of the labels for textual links.
For example, if there are 30 selectable elements on
a web page, it is possible to relabel elements that
have the same starting letter(s) as other elements to
maximize selection efficiency. This approach is sim-
ilar to automatic access keys, and thus faces similar
problems in that the relabeling of many elements con-
tributes to visual clutter or causes layout changes. In
our implementation, the aggressiveness of the label
builder will most likely be tunable based on the input
characteristics of the user. If a user has a very dif-
ficult time activating their input device, efficiency is
of paramount importance, and elements should be re-
labeled for maximum coverage of the alphanumeric
character set. In other cases, minimizing the changes
to the existing labels may be desired to minimize the
impact on the look of the web page.
3.3 Implementation
One of the reasons that previous attempts at custom
browsers for low bandwidth input have been less use-
ful than theorized is that implementations were often
based around custom written HTML rendering en-
gines. Although this approach gives absolute con-
trol over the presentation of web pages, it limits any
implementation’s usefulness for todays highly com-
plex web pages. Fortunately for future developers
in this area, two mature open source HTML render-
ing engines are now available for use. Both Webkit
(webkit.org, 2006) (the core of Apple’s Safari Web
Browser) and Gecko (mozilla.org, 2006b) (the foun-
dation of the Mozilla Firefox browser) can be em-
bedded into custom applications or extended in any
fashion. We are in the process of implementing our
navigation technique as an extension to the Mozilla
Firefox browser.
One obstacle for users not able to move the mouse
is determining the location of selectable elements on
the page. For regular textual links, identification is
not a problem as these links are usually underlined
and colored blue. However, an image link or the links
within an image map may be difficult to locate with-
out having the visual feedback of the mouse cursor
turning to a hand symbol. To overcome this prob-
lem, our implementation will darken any areas of the
page that are not selectable when the link function
is selected. Figure 4 shows an example page with
non-selectable elements darkened. In the regular view
(Figure 3) it is not possible to determine that the im-
age of the stick figure is actually a link. This tech-
nique gives users a visual cue to discover selectable
elements without introducing complexity to the user
interface or modifying the page layout.
Additionally,in Figure 4 the label overlays are vis-
ible for the image link and the text entry box. These
two elements normally do not show a user visible tex-
tual description, so our system assigns a label based
on the information available about the element. In the
case of the image, the system has extracted the al-
ternate text specified in the HTML code and turned it
into a label for the element. If no alternate text is spec-
ified, the system would generate a label consisting of
an unused (as the first character in another label) al-
EFFICIENT NAVIGATION ON THE WORLD WIDE WEB FOR THE PHYSICALLY DISABLED
325
Figure 3: Example Page Containing Different Selectable El-
ements.
Figure 4: Example Page Showing Selective Darkening and
Label Overlays.
phanumeric character and the type of the element (im-
age, check box, text box, etc.). The search query box,
for example, has a label that was constructed in this
manner.
4 TYPING INTERFACE
The appropriate typing interface is highly dependent
on the specific input device and performance char-
acteristics of the user. For single switch users, this
may be a row-column scanning keyboard or Morse
code, while for eye or head tracking users this will be
an on-screen virtual keyboard. Some commercially
available alternative input devices are paired with ap-
propriate software for the user to type into any third
party application. While future developments may re-
quire a more integrated typing interface (Section 6),
initially we intend to leverage existing text entry so-
lutions.
5 DISCUSSION AND
CONCLUSIONS
This paper has presented a new approach to navigat-
ing on the web for people not able to easily use a stan-
dard computer mouse or keyboard. Using text input
to search the labels of selectable elements on a given
page and selecting matching elements, users are able
to acquire and activate their desired element in a pre-
dictable and efficient way. Compared to the scanning
selection technique, our approach improveson perfor-
mance for web pages with many selectable elements,
reduces the time cost of incorrect selections, and in-
creases the system’s predictability for the user. It is
our hope that more efficient tools for web access will
make it feasible for users with severe disabilities to
take full advantage of the world wide web.
Although we believe that the incremental search
selection process will be significantly more efficient
than a linear scanning approach for most web pages,
there are certain cases where it is not optimal. The
most common case is when the web page contains
only a small number of selectable elements. In this
case, the system should fall back to a linear scanning
approach. The exact number of selectable elements
where a linear scanning approach is more efficient de-
pends on the distribution of starting letters in the se-
lectable elements and the characteristics of the virtual
keyboard (which letters are fastest to type). This num-
ber may be calculated independently for each page
based on these properties.
It remains to be seen whether users are more or
less comfortable with the search based selection pro-
cess versus the scanning based one. Although speci-
fying one’s desired target by typing part of its name
seems intuitive, it is a departure from the direct se-
lection modes the user may be familiar with such as
“point and click” or automatic scanning. After the
implementation phase is complete, this issue will be
investigated further through user trials.
A beneficial side effect of the search selection
method is that it gives the user more control over the
process of navigation. With previous approaches, the
system controlled the pace of the scanning and the
user was merely reacting to events instead of control-
ling them. Although the scanning pace can be ad-
justed, these systems have long scanning processes
that require the user’s complete attention and cannot
be interrupted. Not being in control of the system’s
operation can cause users to feel “pushed” and anx-
ious about using the software.
6 FUTURE WORK
On many web pages, some elements are more fre-
quently used or more important relative to other el-
ements on the page. Important elements may be the
main links in the navigation bar, the email body text
WEBIST 2007 - International Conference on Web Information Systems and Technologies
326
box in a web based email page, or the submit button
for a form. In addition, in most virtual keyboard in-
terfaces some letters are easier and faster to type than
others. For example, the characters closest to the top
left corner are usually the fastest to type in a row-
column scanning keyboard since the scanning starts in
the topmost row and leftmost column. If typing using
Morse code, symbols with shorter dot/dash combina-
tions (e, t, a, i, m, n, etc.) are faster to type. Which
characters are the most efficient depends on the spe-
cific typing interface being used.
In future work, we intend to examine matching the
most important elements in a web page to the charac-
ters that are easiest to type in a given interface. This
requires an algorithm to order the selectable elements
based on their estimated importance. A measure for
the importance of an element could be determined
based on a combination of metrics such as text size
of a link label, the actual text of a button, the name
of an input field, or even a dynamic measure such as
the PageRank (Page et al., 1998) of a link. In addi-
tion, this approach would require tighter integration
between the typing interface and the web browser to
determine the cost of typing each character in the in-
terface.
REFERENCES
gok.ca (Nov 2006). Gnome on-screen keyboard.
http://www.gok.ca.
Hackett, S., Parmanto, B., and Zeng, X. (2004). Acces-
sibility of internet websites through time. In As-
sets ’04: Proceedings of the 6th International ACM
SIGACCESS Conference on Computers and Accessi-
bility, pages 32–39, New York, NY, USA. ACM Press.
Hanson, V. L., Brezin, J. P., Crayne, S., Keates, S., Kjeld-
sen, R., Richards, J. T., Swart, C., and Trewin, S.
(2005). Improving web accessibility through an en-
hanced open-source browser. IBM System Journal,
44(3):573–588.
konqueror.org (Nov 2006). Konqueror - web browser.
http://konqueror.org/features/browser.php.
madentec.com (Nov 2006). Discoverpro.
http://www.madentec.com/products/discoverpro.php.
Mankoff, J., Dey, A., Batra, U., and Moore, M. (2002).
Web accessibility for low bandwidth input. In Assets
’02: Proceedings of the Fifth International ACM Con-
ference on Assistive Technologies, pages 17–24, New
York, NY, USA. ACM Press.
mozilla.org (Nov 2006a). Accessibility features in firefox.
http://www.mozilla.org/access/features.
mozilla.org (Nov 2006b). Mozilla layout engine.
http://www.mozilla.org/newlayout/.
Owens, J. and Keller, S. (2000). Multiweb: Australian con-
tribution to web accessibility. In Proceedings of the
9th Australasian Conference on Information Systems.
Page, L., Brin, S., Motwani, R., and Winograd, T. (1998).
The pagerank citation ranking: Bringing order to the
web. Technical report, Stanford Digital Library Tech-
nologies Project.
Prinz, R., Zeman, P. M., Neville, S., and Livingston,
N. J. (2006). Emulating a single switch input device
through wavelet de-noising of surface EMG signals.
In Proceedings of the 3rd Cambridge Workshop on
Universal Access and Assistive Technology.
Rowell, D., Dalrymple, G. F., and Olsen, J. (1978). UNI-
COM: A universal communication and control system
for the non-verbal motor impaired. SIGCAPH Com-
puters and the Physically Handicapped, (24):56–59.
Stephanidis, C., Paramythis, A., Karagiannidis, C., and Sa-
vidis, A. (1997). Supporting interface adaptation in
the AVANTI web browser.
Steriadis, C. E. and Constantinou, P. (2003). Designing
human-computer interfaces for quadriplegic people.
ACM Transactions on Computer-Human Interaction,
10(2):87–118.
w3.org (Nov 2006). Web accessibility initiative.
http://www.w3.org/WAI/.
webkit.org (Nov 2006). The webkit open source project.
http://webkit.org.
Zhao, L., Yan, J., and Yuan, K. (2005). Mozilla acces-
sibility on unix/linux. In W4A ’05: Proceedings of
the 2005 International Cross-Disciplinary Workshop
on Web Accessibility (W4A), pages 90–98, New York,
NY, USA. ACM Press.
EFFICIENT NAVIGATION ON THE WORLD WIDE WEB FOR THE PHYSICALLY DISABLED
327