Low Cost Mark-Up for Lightweight Semantics
Simon Harper and Sean Bechhofer
sharper,seanb [@cs.man.ac.uk]
Information Management Group, Dept of Computer Science,
University of Manchester, Manchester, UK.
http://augmented.man.ac.uk
Abstract. Visually impaired users are hindered in their efforts to access the
largest repository of electronic information in the world, namely the World Wide
Web (Web). A visually impaired user’s information and presentation require-
ments are different from a sighted user in that they are highly egocentric and
non-visual. These requirements can become problems in that the web is visually-
centric with regard to presentation and information order / layout, this can (and
does) hinder users who need presentation-agnostic access to information. Our ob-
jective is to address these problems by creating usable appropriately ‘displayed’
web pages for use by all users who wish to understand the meaning as opposed to
the presentation and order of the information. We assert that the only way to ac-
complish this is to encode the pages semantic information directly into the page.
And the only way this will occur in the real world is if authors have no ‘semantic
overhead’ when creating these pages. In this paper we describe preliminary work
towards a system to enable just this kind of semantic encoding so that, in effect,
authors get low cost semantics.
1 Introduction
We assert that the most preferential way to enhance visually impaired peoples access to
information on web-pages is to encode the meaning of that information into the specific
web-page it refers to. However, there are problems. Empirical evidence suggests that
authors and designers will not separately create semantic mark up to sit with standard
XHTML1 because they see it as an unnecessary overhead.
Recently, we have seen a movement towards a separation of presentation, metadata
(XHTML), and information. However, this has not been enough to support the unfet-
tered access of visually impaired users. Consider, the excellent ‘CSSZenGarden’ (see
Fig. 1). The site is a model of the state-of-the-art: the application of standards, sepa-
ration of presentation and content, and visually stunning too. But, it is still reasonably
inaccessible to visually impaired people. Inspect the site without an applied stylesheet
(see Fig. 2). Visually impaired users interact with these systems in a ‘serial’ (audio)
manner as opposed to a ’parallel’ (visual) manner. Content is read from top left to bot-
tom right, there is no scanning and progress through information is slow. Given this
interaction paradigm we can see that visually impaired users are still at a disadvantage
because they have no idea items are menus, what the page layout is, what the extent
is. In effect, the implicit meaning contained in the visual presentation (see Fig. 1). is
lost and any possibility of enhanced meaning is also not available as only authoring
concepts (like footnote, heading, leftcolumn) are listed (see Fig. 2).
While authors and content creation engines still create non-standard CSS2 -XHTML
1
Extensible Hypertext Markup Language
2
Cascading Style Sheet
91
Fig. 1. Zen Garden with CSS 83
identifers, they also often compound the problem by using linear paper based (book)
metaphors such as: footer, header, bold, big, etc. This information can in fact be in-
ferred from the coded style and presentation information contained within the CSS.
This means the combination of identifer and presentation information together often
represent a tautology.
Even when authoring concepts do look as though they have a meaning with regard to
the information they are often mixed with un-descriptive qualifiers; and the problem is
again compounded by the lack of an ontology in the event of there actually being some
useful information to reason over. Therefore, the question which we faced and which
this paper is dedicated to answering was this:
How can semantic information be built into general purpose web-pages such
that the information is as accessible to visually impaired users as it is to sighted
users, without compromising the page’s design vision?
We based our question on a set of beliefs thus:
1. Visually impaired surfers need access to the meaning of information to assist in
their cognition, perception, movement around that information, and to assist in the
formulation of their world-view [4, 9]. This is the same for sighted users however
pages are normally created with sighted users in mind.
2. Based on empirical and anecdotal evidence, authors and designers will not suffer a
‘Semantic Overhead’ when building pages.
3. A web page should be thought of an application, comprising functional elements
and presentation / information elements, within an application (the browser).
4. Information should not need to be recreated (i.e. exist as XHTML for humans and
RDF3 for agents) when the intended audience is human. The meaning should be
seamless and be part of the data.
3
Resource Description Framework (Schema)
92
5. If we don’t need to create explicit resources (RDF feeds etc) why should we?
6. Authoring concepts used as presentation identifiers are redundant when used with
CSS as their presentational meaning is implicit in their technical definition.
Fig. 2. Zen Garden without a CSS
This goal and set of beliefs led us to a simple, lightweight, and powerful solution. Cre-
ate a grammar to represent the meaning of data within XHTML meta tags and encoded
it into the data by leveraging the ‘class’ and ‘id’ attributes common to most
XHTML elements. CSS presentation will be unaffected but semantics will be an im-
plicit part of the data as opposed to an explicit duplicate representation (in say RDF(s)
or N3 Notation). To achieve this we combine both XHTML elements that have meaning
or that can be used to accurately infer meaning; and a bespoke grammar developed to
enhance the limited XHTML syntax.
The focus of our system is to represent instances as information enclosed within meta
elements along with concept and property identifiers as part of XHTML meta elements
themselves. These elements can then be related to OWL Lite [8] ontologies defined in
the normal way.
1.1 Synopsis
One of the goals of the Semantic Web vision is to make knowledge accessible to agents
but with a strong human input and benefit. In this framework, our goal is to make
the role of the objects, that support visual accessability through presentation, explic-
itly interpretable by humans (via web browsers) rather than just being visually inter-
pretable. Therefore, it is necessary to associate metadata and semantics with XHTML
objects (machine-readable vs machine-understandable). The rest of the paper can be
summarised as follows:
93
Background We give an overview of how visually impaired people currently interact
with web pages. We describe the problems associated with these methods and give
an overview of current access paradigms and authoring concepts.
Related Work We present a small section on related work to place our contribution in
context.
Low Cost Semantics We describe the concepts, rational, and techniques behind our
system focusing on the XHTML abbrv / acronym elements and the ‘class’
attribute. We show how these are referenced on XHTML pages and how our lightweight
system can contribute to the accessibility of information via lightweight semantics.
Example As a preliminary case study we consider the simple ontology taken from ‘A
Semantic Web Primer’ in an attempt to show how an ontology is represented using
our methodology.
Why Does This Approach Aid Visually Impaired Users? We have identified the prob-
lem and suggested a solution but why do we think this is a useful solution?
Conclusion Finally, we focus on our conclusions from the work undertaken and look
at future work including system evaluations.
2 Background
Access to, and movement around, complex hypermedia environments, of which the web
is the most obvious example, has long been considered an important and major issue
in the Web design and usability field [5, 10]. The commonly used slang phrase ‘surfing
the web’ implies rapid and free access, pointing to its importance among designers and
users alike. It has also been long established [4, 6] that this potentially complex and
difficult access is further complicated, and becomes neither rapid or free, if the user is
visually impaired4 .
2.1 Current Access Paradigms
Visually impaired people usually access Web pages either by using screen readers or
specialist browsers. If the Web pages are properly designed and laid out in a linear fash-
ion, these assistive technologies can work satisfactorily. Some screen readers access the
HTML / XHTML source code rather than solely reading the screen, which enables them
to provide better support. However, not many pages are properly designed; the focus is
usually on the visual presentation which makes audio interaction almost impossible.
Furthermore, chunking the page into several parts and presenting it in a nonlinear fash-
ion is becoming popular which makes the provided functionalities of these assistive
technologies insufficient. There are guidelines to aid the designers in creating accessi-
ble pages [1], unfortunately few designers follow these guidelines and therefore Web
accessibility is still a problem.
Further problems also exist when trying to gain an overview of the page. Some screen
readers, for instance Jaws [11], provide overview information when the user first ac-
cesses a page. This information often includes, for example, the number of headings in
the page based on the “heading” tags in the source code. However, if the page is not
appropriately designed, such information could be misleading.
4
Here used as a general term encompassing the WHO definition of both profoundly blind and
partially sighted individuals [13].
94
2.2 The Problem with Authoring Concepts
Even when XHTML meta elements are used correctly and pages are created to stan-
dards and specifications, poor accessability still persists. We believe this is because
there are common misconceptions about what information is actually required by users.
In our opinion this continued inaccessibility stems from the incorrect use of authoring
concepts within the web-page.
Authoring concepts often hold information about the layout vocabularies used in transcod-
ing and content management systems; but from a visual perspective. In this case, they do
not consider the meaning of the objects in the page framework but are more interested
in how the objects are presented in the Web landscape. The Web landscape is defined as
the combination of the page and the agent (e.g, browser and assistive technologies such
as screen readers). These concepts are more to do with the specific structures that can be
used to define the overall layout of a page including for example, sections, summaries,
abstracts, footers, etc. These constructs are usually implicit in the visual presentation of
the page, and so many authors and transcoding systems seek to explicitly encode them
in the underlying source code (e.g., HTML). However, this kind of terminology is less
useful and therefore inaccessible in any other form of interaction (e.g., audio interaction
through screen readers). Transcoders aim to define a vocabulary that is already widely
used between the designers but not formally explained and defined, that is to say they
try to make the domain knowledge explicit. However, they use the wrong paradigm, that
of the linear and visual layout as opposed to the really useful information – the meaning
of the actual instance of data itself.
Authors and systems need to move away from this paradigm of providing what they
THINK users need and focus on what the creator actually MEANS. In this way visu-
ally impaired users can decide for themselves what is useful, and what is not.
3 Related Work
Adding semantics to an XHTML document is not a new concept. It has been thought
about since the late 1990’s however concrete solutions were proposed as early as 2002.
Tim Berners-Lee proposed embedding XML RDF in HTML documents as part of the
tag project [3], however these documents would not validate as XHTML and so did
not find favour among the community [12]. A version was created that did validate by
the inclusion of a small DTD using XHTML Modularisation. However, this was not
deemed a good solution as unique extensions have to be created on a whim. In fact the
work concluded that the RDF specification specifies how to understand the semantics
(in terms of RDF triples) in an RDF document that contains only RDF, but does not
explain how and when one can extract semantics from documents in other namespaces
which contain embedded RDF. It goes on to say that the XHTML specification explains
how to process XHTML namespace content, but gives no indication about how to pro-
cess embedded RDF information [3]. Other methods have been proposed in which the
object or script elements are used, however, the code becomes unreadable and therefore
less workable although the RDF can be linked to in an external file [14]. The use of
the XHTML link element has also been proposed, however the main problem with
this method is that the RDF is not actually then embedded in the HTML source but
in a separate file [14]. This file is then at the mercy of changes and synchronisation
issues with the original and the amount of work needed to create the resource is the
same as creating two separate and disjoint files – time and effort are not saved. Dan
Connolly proposed a system called HyperRDF in which HTML is used as the conduit
95
to use XSLT to transform information into RDF. However, HyperRDF cannot be vali-
dated since the head element does not allow an ID attribute [7]. Augmented Metadata
for XHTML is an implementation that allows Dublin Core metadata to be incorporated
in Web pages in a way that is compatible with today’s Web browsers. The basic premise
is that one can take the profile attribute to be a global namespace prefix for all of the
rel / meta and name attributes throughout the document. This approach is mainly
for those authors that want to use a simple mechanism for producing RDF from their
XHTML. It is ineffective from the point of view of anyone that wants to randomly ex-
tract RDF from XHTML, since one cannot tell whether the author wanted the assertions
to be converted into the triples produced by the algorithm or not [2].
Finally, the most recent thinking on the subject comes in the form of GRDDL (Gleaning
Resource Descriptions from Dialects of Languages). This work is being undertaken by
the W3C Web Co-ordination Group and is a mechanism for encoding RDF statements
in XHTML and XML. GRDDL shares some commonalities with HyperRDF and works
on the principle that the HTML specification provides a mechanism for authors to use
particular metadata vocabularies and thereby indicate the author’s intent to use those
terms in accordance with the conventions of the community that originated the terms.
Authors may wish to define additional link types not described in this specification. If
they do so, they should use a profile to cite the conventions used to define the link
types. GRDDL is one of these profiles which uses XSLT to transform a page to an RDF
description.
3.1 Why GRDDL Doesn’t Work For Us
Our research centres around both the designer and the user. We wish to support the
designer because in doing this we make sure our target user group are supported by
the designers’ creation. In our conversations with designers the resounding message we
receive is
“If there is any kind of overhead above the normal concept creation then we
are less likely to implement it. If our design is compromised in any way we will
not implement. We create beautiful and effective sites, we’re not information
architects.”
Many web designers move from print media to web design and this pre-gained experi-
ence in creating static designed artifacts forces them to see design as fixed and immov-
able once created. A designer creates and controls the development of what is in effect
a piece of art and therefore once created should not be changed or violated. It can be
difficult to convey that users often require web pages to adapt to their needs, and the
fact that this sometimes goes beyond art.
We suggest that designers need a lightweight no-frills approach to include semantic
information within XHTML documents; in effect the presence of the semantic infor-
mation should be seamless indivisible and have a low cost design overhead.
4 Low Cost Semantics
Our system is in reality a process for associating ontology concepts with instances en-
coded within XHTML pages. Currently, presentation and meaning are separated as we
can see in figure 3. The CSS and ontologies are mostly manual created while the in-
stances, the XHTML, and the semantics associated with instances are created either
96
manually or are automatically generated. The CSS and XHTML are assembled on the
client and joined by the browser functionality while the RDF and ontology are used by
either automated agents or RDF ‘feed’ readers (for use by humans). We suggest that
this type of separation is both unhelpful, damaging, and counter to the Semantic Web
vision. With Tim Berners-Lee’s desire to describe resources (many on the web as stan-
dard XHTML documents) more fully the division between the web and the sematic web
will increasingly become a hindrance. Although users can currently interact with web
resources, and agents are starting to interact with sematic resources, surely progress
should be made towards a joining of the two. We believe there should be just one web
where semantics, presentation, and information are conjoined giving a holistic world-
view.
Our system is a first step towards this. We suggest that meaning should be encoded
within the elements of the XHTML and CSS along with ontologies which can be cre-
ated as normal. Ontological concepts and properties are encoded into both the elements
and attributes of the XHTML document and are used as identifiers within the CSS which
link presentation to XHTML elements. Our system revolves around a software process
(see Fig. 4) which converts an RDF–XHTML document into a series of instances and
ontological descriptors for supply to the reasoner. Users view the document in a web
browser as normal, however, browsers that are ‘semantic-aware’ can use the ontologi-
cal information to provide more intelligent access to the instances of information than
before. Currently, no browsers are ‘semantic-aware’ of our system except those with a
system plug-in. However, all is not lost as RDF(s) can be generated by our process and
inserted into the document such that RDF(s) aware browsers can take advantage of our
system (as a ‘Kludge’).
Fig. 3. As Things Currently Stand Fig. 4. Our Preliminary System
97
4.1 Encoding Ontologies in XHTML
Because we are suggesting a lightweight system our paradigm for encoding OWL Lite
ontologies is simple, flexible, and without a semantic overhead. We use a trinity of
techniques to encode semantics directly into a page:
Class and ID Attributes XHTML class or id attributes are used to encode a piece
of semantic information in the form of a concept-class or property into a defined
piece of XHTML delimited by the closing element identifier. This is normally
achieved by using the div and span elements to conjoin both the presentation
style (CSS) and the semantic meaning (ontology) to the user (see Fig. 6).
Non Presentational XHTML Attributes We can leverage the implicit information con-
tained in the names of XHTML elements if we have a corresponding ontology. Ele-
ments that are non-presentational (like ) can be used to encapsulate
meaning within the page (see Fig. 6).
Individuals Unique individuals are defined by use of the anchor element where the
href attribute is used to point to the URI or MAILTO of the unique item. If http
/ mailto are used then the link will be click-able. If uri is used then the link is
not click-able (see Fig. 6).
Fig. 5. RDF and RDFS Layers Taken from ‘A Semantic Web Primer’ Pg 84
We include namespaces in XHTML documents so that multiple ontologies can be used
to describe one document. To implement this we use the link element of the XHTML
header section.
98
The first line represents the format the second an example. The rel attribute is always
‘ontology’ as this differentiates it from stylesheets and the like. Elements can be related
to a namespace by using either the namespace identifier in the class attribute of the
enclosing div element or by joining the namespace to the attribute name using an un-
derline ( ). The suggested approach provides a mechanism for encoding ”lightweight”
information. Of course this approach has its limitations – we can capture simple instan-
tiation of atomic classes along with property assertions, but not richer assertions such
as instantiation of arbitrary class expressions. We stress that this is not intended as a
replacement for other representations but is a complementary mechanism. For exam-
ple, we can still expect the class and property definitions in the ontology to be encoded
using existing approaches such as RDF/XML.
Designers often want to adjust the visual design of a web-page without altering the
actual meaning. We support theorise that this ad-hoc visualization can be handled by
specialising ontological concepts with visual extensions if required.
5 Example
As a preliminary case study let us consider the simple ontology taken from ‘A Semantic
Web Primer’ (in press) page 84 Figure 3.6 and recreated here for convenience as Fig. 5.
Let us now see how information culled from ‘David Billington’s’ Web page can be
annotated (see Fig. 6) such that the semantics of the instance are available for inference
following the ontology in Fig. 6. We can see that this information is just a general
description of the course information. However, by adding a div element we enclose
the information such that the enclosure implicitly relates any enclosed sub-elements.
Secondly, we see that a span and anchor element are introduced to denote Course and
IsTaughtBy. We can via the ontology now infer the conceptual range (using ABox
reasoning via ‘Racer’) that discrete mathematics is taught by the associate professor
David Billington, and what is more, so can assistive agents. This seems to represent
what we want to say from a reasoning approach and when presented it is displayed
correctly and with no additional overhead for the designer.
Discrete Mathematics, taught by
David Billington, is a second year course designed for Computer Science
students who need a more formal mathematics training.
Fig. 6. XHTML Code
99
6 Why Does This Approach Aid Visually Impaired Users?
By knowing the meaning of the information that is being encountered visually im-
paired users can perform their own triage on that information. As we have previously
mentioned, web pages are read from top left to bottom right. If there is a lot of infor-
mation on the the page then the user can get lost, disoriented, or at least frustrated with
their progress through this information. By presenting the meaning of the information
using standard transcoding methods, users can choose which information is important
to them, not the visual designer.
7 Conclusions
Our system suggests a method of encoding lightweight mark-up into webpages to incur
a low cost semantic benefit. With the meat of the information design being abstracted
from the graphic / web designer the system has given a taster of how semantics can
be represented within web-pages. Additionally. we also show how this can be achieved
without incurring a significant overhead with regard to marking-up that semantic in-
formation and have it validate to XHTML 1.0 strict.. We propose that the inclusion
of semantic information directly into the XHTML is the only way to assist visually
impaired users access web pages while not increasing or compromising the creation
activity of authors and designers. Indeed we show the first stage in a more elaborate
system to enable semantic information to be freely accessible by all users.
References
1. Web content accessibility guidelines 1.0, 1999. http://www.w3.org/TR/1999/WAI-
WEBCONTENT/.
2. M. Altheim and S. B. Palmer. Augmented Metadata in XHTML, 2002. http://
infomesh.net/2002/augmeta/ - valid 2004.
3. T. Berners-Lee. RDF in HTML, 2002. http://www.w3.org/2002/04/htmlrdf -
valid 2004.
4. M. Brambring. Mobility and orientation processes of the blind. In D. H. Warren and E. R.
Strelow, editors, Electronic Spatial Sensing for the Blind, pages 493–508, USA, 1984. Dor-
drecht, Lancaster, Nijhoff.
5. C. Chen. Structuring and visualising the www by generalised similarity analysis. In Pro-
ceedings of the 8th ACM Conference on Hypertext and Hypermedia, New York, USA, 1997.
ACM Press.
6. A. Chieko and C. Lewis. Home page reader: IBM’s talking web browser. In Closing the Gap
Conference Proceedings, 1998.
7. D. Connolly. HyperRDF: Using XHTML Authoring Tools with XSLT to produce RDF
Schemas, 2000. http://www.w3.org/2000/07/hs78/ - valid 2004.
8. M. Dean and G. Schreiber. OWL Web Ontology Language Reference. W3C Recommenda-
tion, World Wide Web Consortium, 2004. http://www.w3.org/TR/owl-ref/.
9. A. G. Dodds. The mental maps of the blind. 76:5–12, January 1982.
10. R. Furuta. Hypertext paths and the www: Experiences with walden’s paths. In Proceedings
of the 8th ACM Conference on Hypertext and Hypermedia, New York, USA, 1997. ACM
Press.
11. Henter-Joyce, Inc. Jaws. http://www.hj.com.
12. N. Kew. Why Validate?, 2002. http://lists.w3.org/Archives/Public/
www-validator/2001Sep/0126.html - valid 2004.
13. V. RNIB. A short guide to blindness. Booklet, Feb 1996. http://www.rnib.org.uk.
14. Sean B. Palmer. RDF in HTML: Approaches, 2002. http://infomesh.net/2002/
rdfinhtml/ - valid 2004.
100