Issues in Informing Science and Information Technology Volume 4, 2007

Automatic Conceptual Analysis for

Plagiarism Detection
Heinz Dreher
School of Information Systems, Curtin University of Technology
Perth, Western Australia
h.dreher@curtin.edu.au

Abstract
In order to detect plagiarism, comparisons must be made between a target document (the suspect)
and reference documents. Numerous automated systems exist which check at the text-string level.
If the scope is kept constrained, as for example in within-cohort plagiarism checking, then per-
formance is very reasonable. On the other hand if one extends the focus to a very large corpus
such as the WWW then performance can be reduced to an impracticable level. The three case
studies presented in this paper give insight into the text-string comparators, whilst the third case
study considers the very new and promising conceptual analysis approach to plagiarism detection
which is now made achievable by the very computationally efficient Normalised Word Vector
algorithm. The paper concludes with a caution on the use of high-tech in the absence of high-
touch.
Keywords: academic malpractice, conceptual analysis, conceptual footprint, semantic footprint,
Normalised Word Vector, NWV, plagiarism.

Introduction
Plagiarism is now acknowledged to pose a significant threat to academic integrity. There is a
growing array of software packages to help address the problem. Most of these offer a string-of-
text comparison. New to emerge are software packages and services to ‘generate’ assignments.
Naturally there will be a cat and mouse game for a while and in the meantime academics need to
be alert to the possibilities of academic malpractice via plagiarism and adopt appropriate and
promising counter-measures, including the newly emerging algorithms to do fast conceptual
analysis. One such emergent agent is the Normalised Word Vector (NWV) algorithm (Williams,
2006), which was originally developed for use in the Automated Essay Grading (AEG) domain.
AEG is a relatively new technology which aims to score or grade essays at the level of expert
humans. This is achieved by creating a mathematical representation of the semantic information
in addition to checking spelling, grammar, and other more usual parameters associated with essay
assessment. The mathematical represen-
Material published as part of this publication, either on-line or
in print, is copyrighted by the Informing Science Institute.
tation is computed for each student es-
Permission to make digital or paper copy of part or all of these say and compared with a mathematical
works for personal or classroom use is granted without fee representation computed for the model
provided that the copies are not made or distributed for profit answer. If we can represent the semantic
or commercial advantage AND that copies 1) bear this notice content of an essay we are able to com-
in full and 2) give the full citation on the first page. It is per-
missible to abstract these works so long as credit is given. To pare it to some standard model- hence
copy in all other cases or to republish or to post on a server or determine a grade or assign an authen-
to redistribute to lists requires specific permission and payment ticity parameter relative to any given
of a fee. Contact Publisher@InformingScience.org to request
redistribution permission.
Automatic Conceptual Analysis for Plagiarism Detection

corpus; and create a persistent digital representation of the essay.

AEG technology can be used for plagiarism detection because it processes the semantic informa-
tion of student essays and creates a semantic footprint. Once a mathematical representation for all
or parts of an essay is created it can be efficiently compared to other similarly constructed repre-
sentations and facilitate plagiarism checking through semantic footprint comparison.

The Plagiarism Problem

The extent of plagiarism is indeed significant. Maurer et al. (2006) provide a thorough analysis of
the plagiarism problem and possible solutions as they pertain to academia. They divide the solu-
tion strategies into three main categories. The most common method is based on document com-
parison in which a word for word check is made with each target document in a selected which
could be the source of the copied material. Clearly this is language independent as one is essen-
tially comparing character strings; it will also match misspellings. The selected set of document is
usually all documents comprising assignment or paper submissions for a specific purpose. A sec-
ond category is an expansion of the document check but where the set of target documents is
‘everything’ that is reachable on the internet and the candidate to be checked for is a characteris-
tic paragraph or sentence rather than the entire document. The emergence of tools such as Google
has made this type of check feasible. The third category mentioned by Maurer et al. is the use of
stylometry, in which a language analysis algorithm compares the style of successive paragraphs
and reports if a style change has occurred. This can be extended to analyzing prior documents by
the same author and comparing the stylistic parameters of a succession of documents.
However, the issue of plagiarism is not merely a matter for academics. Austrian journalist Josef
Karner (2001) writes “Das Abschreiben ist der eigentliche Beruf des Dichters” (“Transcription is
the virtual vocation of the poet”). Is then the poet essentially a professional plagiarist, taking oth-
ers’ ideas and presenting them in verse as his own and without attribution? This may be a rather
extreme position to hold, but its consideration does point up interesting possibilities which the
etymology of plagiarism may illuminate.
As yet there is a paucity of statistics available to help us understand the extent of plagiarism.
However a recent Canadian study (Kloda & Nicholson, 2005) has reported that one in three stu-
dents admit to turning to plagiarism prior to graduation - serious enough one may think. The truly
shocking statistic is that one in 20 has actually paid for someone else to write or provide an as-
sessment paper which they subsequently submitted as their own (Figure 1).
# of Canadian students who admit to plagiarizing at least once before graduating = 1/3
# of Canadian students who admit to submitting a paper they had purchased
online as their own = 1/20
# of Canadian universities and colleges that subscribe to Turnitin™ = 28/90 (31%) or 28/130 (22%)
Turnitin™ is one of the well established plagiarism checking systems

Figure 1: Plagiarism statistics (Kloda & Nicholson, 2005)

Despite having plagiarism detection technology available, its effective implementation can be a
challenge in itself. In one prominent case where technology was forced on students, the reaction
led to a court ruling granting the student the right to bypass a university mandated plagiarism
check prior to assignment submission (Figure 2).

602
 Dreher

A student at McGill University has won the right to have his assignments marked without first submit-
ting them to an American, anti-plagiarism website.
Jesse Rosenfeld refused to submit three assignments for his second-year economics class to Tur-
nitin.com, a website that compares submitted works to other student essays in its database, as well as
to documents on the web and published research papers.
Last Updated: Friday, January 16, 2004 | 11:11 AM ET
Figure 2: McGill student wins fight over anti-cheating website
Source: http://www.cbc.ca/canada/story/2004/01/16/mcgill_turnitin030116.html

To help effective plagiarism detection implementation in educational institutions around the

world, advice for students and staff is readily available on a growing number of plagiarism-
dedicated web-based resources – a sample appears in Figure 3.
Australia http://www.lc.unsw.edu.au/onlib/plag.html
http://academichonesty.unimelb.edu.au/
http://startup.curtin.edu.au/study/plagiarism.html
http://www.teachers.ash.org.au/aussieed/research_plagiarism.htm
Austria http://www.iaik.tugraz.at/aboutus/people/poschkc/EinfuehrungInDieTelematik.htm
http://ipaweb.imw.tuwien.ac.at/bt/index.php?id=aktuelles
Canada http://library.acadiau.ca/tutorials/plagiarism/
http://www.library.ualberta.ca/guides/plagiarism/
http://www.ucalgary.ca/~hexham/study/plag.html
Germany http://plagiat.fhtw-berlin.de/
http://www.frank-schaetzlein.de/biblio/plagiat.htm
http://www.spiegel.de/unispiegel/studium/0,1518,227828,00.html
USA http://plagiarism.phys.virginia.edu/ ;
http://www.umuc.edu/distance/odell/cip/links_plagiarism.html
http://www.ece.cmu.edu/~ee240/)
Figure 3: Sample plagiarism resources

Whilst the plagiarism problem is significant, it is not solvable only by applying plagiarism detec-
tion techniques. There needs to be a recognition that the students are not entirely to blame (Wil-
liamsJ 2002). Quite obviously we need to agree on a working definition of plagiarism which is
simple to understand and to check.
In a light-hearted vein, the entry for plagiarism in The Devil’s Dictionary by Ambrose Bierce
reads: PLAGIARISM, n.
A literary coincidence compounded of a discreditable priority and an honorable subsequence.

This might be the sort of definition which would be used to justify excusing a first or minor in-
stance of plagiarism but it does not admit of the measures which may be needed to detect it. A
more precise and practically applicable definition, that indicates the measures which may be
needed to detect plagiarism, is found on the www.plagiarism.org site:
• copying words or ideas from someone else without giving credit
• changing words but copying the sentence structure of a source
without giving credit
• copying so many words or ideas from a source that it makes up the majority
of your work, whether you give credit or not (see our section on "fair use" rules)

From the above we can see the essential elements: words; style or structure; and ideas. Therefore,
checking systems must look for matching words, analyze style, and create a map of the ideas con-

603
Automatic Conceptual Analysis for Plagiarism Detection

tained in candidate plagiarism cases. The first of these is well catered for by the established sys-
tems, such as string-of-text matching.

Established Plagiarism Checkers

As can be seen from Maurer et al. (2006), there are many systems doing the string-of-text match-
ing. Here we briefly consider the performance of two of them which were readily available to the
author – WCopyfind and EVE2.

Case 1: WCopyfind
The University of Virginia’s freely available WCopyfind software
(http://plagiarism.phys.virginia.edu) is a delightful example of the power of the computer to help
in addressing the plagiarism problem. It makes text-string comparisons and can be instructed to
find sub-string matches of given length and similarity characteristics. Such fine tuning permits the
exclusion of obvious non-plagiarism cases despite text-string matches.
To determine the efficacy of WCopyfind the author devised a trial. Some 600 student assign-
ments from a course on Societal Impacts of Information Technology were checked for within-
cohort plagiarism. The assignments were between 500 and 2000 words and were either in the
English or German language. The system is computationally very efficient and took only seconds
to highlight five cases requiring closer scrutiny.
Figure 4, Figure 5, and Figure 6 show WCopyfind – system interface, WCopyfind – report, and
WCopyfind – document comparison, respectively.

Figure 4: WCopyfind – system interface

Since the WCopyfind works at the string-of-text level, language is unimportant and matches are
readily identified from the candidate documents submitted for analysis. Note that such a proce-
dure cannot find plagiarism based on documents not submitted, for example Web resident docu-
ments. Of course, further analysis of a small subset can be submitted for Web-based document
comparison with Google for example. In this case a sample of the identified within-cohort plagia-

604
 Dreher

rized text was submitted for a Google search and immediately revealed a source on the Web con-
taining the same text (Figure 7).

Figure 5: WCopyfind – report

Figure 6: WCopyfind – document comparison

605
Automatic Conceptual Analysis for Plagiarism Detection

It is interesting but perhaps not surprising to note that those who plagiarize from fellow students
will also copy from elsewhere (personal experience). The analysis thus far has not proven plagia-
rism but simply highlighted its possible existence and located the evidence. Simply because text-
strings match does not permit one to conclude plagiarism, as the text may be properly referenced.
The suspect text was found in document www.bsi.de/fachthem/rfid/RIKCHA.pdf (Figure 7) and
can now be carefully matched with student text to determine the extent and accuracy of the copy-
ing. In short, WCopyfind is one text-string-matching approach to plagiarism detection that is use-
ful for within-cohort applications, but is not amenable to large scale ‘extra-cohort’ plagiarism
detection (i.e., searching the WWW). Case study 2 investigates one program that is designed for
this purpose.

Figure 7: Google finds matching Web document

Case 2: EVE2
The Essay Verification Engine (EVE2) by CaNexus.com makes a reliable check of the Internet to
track down possible instances of plagiarism. It examines the essays, and then quickly makes a
large number of searches of the Internet to locate suspect sites. EVE2 visits each of these suspect
sites to determine if they contain work that matches the essay in question. It all sounds rather
simple and straightforward and in a testimonial one reads: “...EVE aced the test, finding every-
thing I had plagiarized. EVE is faster, testing four papers in fifteen minutes, a fraction of the
four hours it took Plagiarism.org to respond.” (excerpted and adapted from the EVE2 website
http://www.canexus.com/eve/).
Naturally such claims are encouraging but will it really work so well in my case? That’s the real
question. Given the ‘speed’ claim above the author decided to submit 16 pages comprising some
7,300 words of a Master’s thesis which was in preparation. It was a chapter on “previous work
and literature review” so one would expect to find some matches.
The first observation was that EVE2 took circa 20 minutes to complete the task. This is not fast at
all; rather, it is so slow that one could not check but a few carefully selected items at this rate.
Figure 8 shows the computer’s CPU usage during the analysis.

606
 Dreher

Figure 8: EVE2 computational demand

The result was ‘disappointing’ too – EVE2 only flagged a low level of potential plagiarism most
of which was due to legitimate referencing and flagged two websites (Figure 9). On the other
hand one is delighted that one’s research students are creating their own work!

Figure 9: EVE2 result for Master’s thesis chapter

The Limitation of String-Of-Text Plagiarism Checking

Finding exact matches of text is, in principle, a simple and straightforward matter, and having
computer support certainly makes it possible to undertake checking on a large scale. However
one usually does not want everything checked against everything else since there will always be
some legitimate text matches. In the two case analyses above the author has found that checking
at the level of text-strings can be useful, especially as the process is language independent (al-
though translations of plagiarized text will not be revealed), and very fast if the scope is restricted
(Case 1).
In Case 2 the processing required to check thousands of words in the source with billions+ of
documents on the web is obviously impractical in the usual situation and would have to be re-
served for special cases where the outcome is critical.
Typically however, further and usually laborious analysis and detection will be required and hu-
man intervention and attention cannot be spared. There are some cases which are rather obvious

607
Automatic Conceptual Analysis for Plagiarism Detection

to humans but not so simply detected automatically. Consider the assignment fragment in Figure
10. These words appeared in an assignment submitted by a student doing a capstone course in
Information Systems & Technology.
Web sites involves a mixture development between print publishing and software development, be-
tween marketing and computing, between internal communications and external relations, and be-
tween art and technology. Software engineering provides processes that are useful in developing the
web sites and web site engineering priniciples can be used to help bring web projects under control
and minimize the risk of a project being delivered late or over budget.

Figure 10: Excerpt from student essay (coursework degree)

As one reads the words one senses a rather unsophisticated level of expression up until the first
comma, then the reader is suddenly confronted with a rather engaging triplet of comparisons to
‘complete’ the sentence. The second and final sentence reverts to the banality with which the
paragraph began.
Looking at the context of the triplet of comparisons (“between marketing and computing, be-
tween internal communications and external relations, and between art and technology”) from
Figure 10 it can be seen that these words do not really integrate into the sentence naturally –
somehow it seems contrived. Such an instance can spark further investigation and in this case led
to the identification of a published article which used these very words. Interestingly, and not at
all surprisingly, this article was the source of considerable chunks of un-attributed replicated text
in the student assignment.
Does automated support for more sophisticated processing of essays, perhaps at the semantic
level exist and which may complement the string-of-text analyses considered above?
Before turning our attention to this, one should mention here that of the three approaches to the
plagiarism problem identified at the outset – namely: words; style or structure; and ideas – the
author has covered the first, and merely makes a reference to the possibility of style analysis,
leaving its treatment to a future paper.
Turning our attention now to the type of processing done by humans, i.e. idea processing, we con-
sider applying the new and very computationally efficient Normalised Word Vector (NWV) algo-
rithm (WilliamsR, 2006) to the task of plagiarism detection.

Conceptual Analysis for Essay Grading

Considerable knowledge and intelligence is needed to detect plagiarism at the concept or idea
level irrespective of which actual words are used to express the idea. Humans are able to do this
very well of course, naturally one might say, but are limited severely by capacity. Computers can
find text-string matches but cannot readily prioritize the cases for manual checking (this is a well
known problem in searching – one receives millions of hits within fractions of a second), which
clearly leaves a gap to be filled by some more sophisticated technology.
The NWV algorithm was devised by Bob Williams for the Automated Essay Grading system
MarkIT developed by Bob Williams and Heinz Dreher at Curtin University of Technology in
Perth, Western Australia (www.essaygrading.com). In essence the approach is to use vector alge-
bra to represent similarities in content between documents. A thesaurus is used to normalise
words by reducing selected words from essay to a thesaurus root word, and deriving occurrence
frequency measures to create a vector representation. This mathematical representation is a con-
ceptual footprint of the essay and can be used for comparison purposes including plagiarism
checking at the semantic level.

608
 Dreher

The next section of the paper presents some case analyses using a promising new technology to
aid in plagiarism detection – the use of the Normalised Word Vector (NWV) algorithm to create a
conceptual footprint of student assignments.

Case 3: Conceptual Analysis Using NWV

In this case study the author has selected the 51 English language essays from the set used in Case
1. Since the thesaurus currently being used by the NWV is an English language thesaurus it fol-
lows that only English language essays can be processed.
Firstly, note that the time taken to process these 51 essays, which is to do a conceptual analysis,
was three minutes (Figure 11).

Figure 11: Concept analysis of 51 essays

Next, look at the conceptual footprint matching, by considering a comparison of an assignment
entity X with itself. In this case X = assignment M0130097tr written on a topic in computer secu-
rity. Inspection of Figure 12 reveals the desired, and expected, perfect match on a concept by
concept basis. Note how the un-normalised words highlighted in the text (right panel, upper com-
pared with middle) match and how the bar graphs match in the lower panel). The left side of the
screen provides a map of the objects being conceptually compared – document files containing
the assignments, and paragraphs within documents, with matching %closeness. The object docu-
ment under consideration obviously has 100% closeness.

609
Automatic Conceptual Analysis for Plagiarism Detection

Figure 12: Conceptual analysis of X with X – 100% closeness

The user, the plagiarist detective or plagiarism analyst, can focus on particular concepts (with
matching thesaurus root definitions). Observe that using this visual feedback one could engage in
quite a meaningful and persuasive discussion with a ‘plagiarist’ for the purpose of establishing
the presence and level of copying.
A focus on concept by concept reveals the lowest level of granularity available with our system.
In the example shown in Figure 13 the concept with name “hiding” is being scrutinized. The the-
saurus entry is shown on the left and any words belonging to the thesaurus entry for “hiding”,
which appear in either of the documents of interest, are shown in the upper right panels.
Obviously such detailed scrutiny even with automated support is time consuming and would be
reserved for special and few cases of intense interest or where the stakes are high.
Dynamically generated maps showing a visual or graphical representation of the conceptual foot-
print and facilitates plagiarism analysis and discussion.
At this juncture our development work has produced a report showing the results of the “Extract
from master essay” components juxtaposed with conceptually similar “Matched extract(s)”. Such
reports are amenable for use in checking large numbers of essays for possible similarity at the
semantic level. An extract from a sample report is shown in Figure 14. For a large number of
essays such a report would contain potentially hundreds of pages if the conceptual similarity
among the essays is significant. With regard to the entire process described, the report production
and reading consumes the greatest resource for computer and human respectively.

610
 Dreher

Figure 13: Analysis of concept with name “hiding”

Figure 14: Extract from a sample conceptual analysis report

Since conceptual analysis is the main topic being addressed in this paper we consider a second
example – this time with assignments of 300 to 500 words written by year 10 students on the
topic of School Leaving Age” (Figure 15).

611
Automatic Conceptual Analysis for Plagiarism Detection

Master essay: B18 E4129 50G.txt

The age at which students are legally allowed to leave school should not be raised from 15 years of age
to 17 years of age.
If a student is kept in school against their will they are less likely to do well in their studies. As a major-
ity of them wont even try to do well and learn what is being tort. They will also disrupt other student that
are willing to learn.
Also at the age of 15 these people are becoming adults. They are beginning to form their own ideas of
what they want to do in the future. Some students are good at school and it would benefit them to stay
others are not good at school and are better off in the work force. These people know there strengths
and weaknesses and are there for better equipt to make the decision for them selves.
People have to be allowed to make their own mistakes. There are always options if the person is un-
happy with their decision. Such as seeking higher education in the field they wish to go into at places like
TAFE or they could even go back to school if they decide they have made the wrong choice. The point is
it should be their responsibility to make the choice that is going to effect the rest of there lives.

Plagiarism detected
Extract from the master essay

Paragraph 1

The age at which students are legally allowed to leave school should not be raised from 15 years of age
to 17 years of age.

Matched extract(s)

Target essay: B14 E4137 117R.txt, paragraph 1, closeness 50%

I belivev that the legal age to leave school should be raise from the age of 15 to the age of 17.

Target essay: B14 E4137 182G.txt, paragraph 1, closeness 63%

I agree with the Minister of Education, that the legal age that students should be allowed to leave school,
is at the age of 17 years old not 15 years old.

Target essay: B20 E4129 179G.txt, paragraph 3, closeness 50%

The arguement that is stated in this essay is should students be allowed to leave school at 15 years of
age or should be change to a later age.

Target essay: B21 E4129 155G.txt, paragraph 1, closeness 57%

Whether you student have to atend school at 15 to 17 years of age.

Target essay: B3 E4126 268G.txt, paragraph 5, closeness 77%

* According to the Minister of Education, the legal age for students to leave school will be changed from
15 years of age to 17 in 2002.

Figure 15: Concept analysis – “School Leaving Age” example

The “Master essay” is presented in full at the beginning of the report. Then for each paragraph of
the “Master essay” (labeled Paragraph 1 in the yellow/light panel) the “Matched extracts” from
other essays (five instances of Target essay in the turquoise/darker panels) are presented with an
indication of the conceptual closeness as a percentage agreement with the “Master essay”. In the
example, five essays contain conceptually similar content. Of course this does not necessarily

612
 Dreher

indicate plagiarism, however the reader may contemplate how well the computer, the NWV algo-
rithm actually, has determined semantic proximity, which may be an indicator of plagiarism.

Conclusion
Through three case studies the author has illustrated how text-string comparison can be effective
in detecting within-cohort plagiarism (Case 1), but can be inefficient for plagiarism detection on a
lar-ger scale such as the WWW (Case 2). Furthermore it has been shown that while text-string
comparisons are effective they may not flag the replication of others’ ideas using semantically
similar words. To detect such forms of copying one needs to use a conceptual analysis. We have
applied the NWV algorithm because it is the fastest method known to extract semantic content
from essays of arbitrary length, the efficacy of which was shown in Case 3.
Whilst the results achieved with this ‘hi-tech’ approach are promising one should stress that a ‘hi-
touch’ approach is not to be ruled out and may be used in a complimentary manner for increased
efficacy in detecting and addressing plagiarism (Figure 16).
In Fig 16 the ‘hi-tech’ approach can be seen used in step 2) whereas the ‘hi-touch’ approach is
relied upon for the remainder of the steps. The term ‘hi-tech/hi-touch’ comes from Naisbitt
(1982). As in all cases where humans rely on technology to help solve problems, in this situation
there is a very large degree of reliance on human (6 out of 7 steps), as opposed to artificial, intel-
ligence.
1) select some text fragment which is ‘unlikely’ to come from the nominated source and search for
<selected text>
2) compare search results with original & highlight matching text
3) professor invites student for interview – bring paper copy of assignment
4) ask student to highlight all words which have been copied
5) compare student’s highlighting with professor’s highlighting and you can guess the student’s
reaction: DISBELIEF
6) professor listens patiently student’s explanation, protestation, justification .
7) professor explains:
if we are HONEST in the assessment process and with each other,
then we can TRUST that the system is FAIR to everyone
and society will RESPECT the worth of your degree from this university:
for this reason we both have the RESPONSIBILITY to uphold academic INTEGRITY

Figure 16: Low-Tech & High-Touch Plagiarism Detection Method

References
Bierce, A. (1911). The Devil’s dictionary. Retrieved from http://www.thedevilsdictionary.com - text by
Ambrose Bierce, 1911; copyright expired.
Kloda, L.A. & Nicholson, K. (2005). Plagiarism detection software and academic integrity: The Canadian
perspective. In Proceedings Librarians’ Information Literacy Annual Conference (LILAC), London
(UK). Retrieved from http://eprints.rclis.org/archive/00005409/
Karner, J. (2001). Der Plagiator Retrieved from http://old.onlinejournalismus.de/meinung/plagiator.html
Maurer, H., Kappe, F. & Zaka, B. (2006). Plagiarism – A survey. Journal of Universal Computer Science,
12(8), 1050-1084.
Naisbitt, J. (1982). Megatrends. Ten new directions transforming our lives. Warner Books.
Turnitin. (2007). http://www.turnitin.com/static/home.html

613
Automatic Conceptual Analysis for Plagiarism Detection

Williams, J.B. (2002). “The plagiarism problem: Are students entirely to blame?” In Proceedings of AS-
CILITE 2002. Retrieved from
http://www.ascilite.org.au/conferences/auckland02/proceedings/papers/189.pdf
Williams, R. (2006). The power of normalised word vectors for automatically grading essays. The Journal
of Issues in Informing Science and Information Technology, 3, 721-730. Retrieved from
http://informingscience.org/proceedings/InSITE2006/IISITWill155.pdf

Biography
Heinz Dreher is Associate Professor in Information Systems at the
Curtin Business School, Curtin University, Perth, Western Australia.
He has published in the educational technology and information sys-
tems domain through conferences, journals, invited talks and seminars;
is currently the holder of Australian National Competitive Grant fund-
ing for a 4-year E-Learning project and a 4-year project on Automated
Essay Grading technology development, trial usage and evaluation; has
received numerous industry grants for investigating hypertext based
systems in training and business scenarios; and is an experienced and
accomplished teacher, receiving awards for his work in cross-cultural
awareness and course design. In 2004 he was appointed Adjunct Pro-
fessor for Computer Science at TU Graz, and continues to collaborate in teaching & learning and
research projects with European partners.
Dr Dreher’s research and development in the hypertext domain has centred on the empowering
aspects of text & document technology since 1988. The systems he has developed provide sup-
port for educators and teachers, and document creators and users from business and government.
‘DriveSafe’, ‘Active Writing’, ‘The Effectiveness of Hypertext to Support Quality Improvement’,
‘Water Bill 1990 Hypertext Project’, ‘A Prototype Hypertext Operating Manual for LNG Plant
Dehydration Unit’, ‘Hypertextual Tender Submission - Telecom Training Programme’, were all
hypertext construction and evaluation projects in industry or education. The Hypertext Research
Laboratory, whose aim was to facilitate the application of hypertext-based technology in
academe, business and in the wider community, was founded by him in December 1989

Acknowledgements
The author would like to acknowledge the InSITE reviewers for their helpful comments and in
particular thank Carl Dreher for his extensive and critical appraisal of early drafts of the paper.

614