Writing Aid Dutch
Supporting Students’ Writing Skills by Means of a String and Pattern Matching
based Web Application
Margot D’Hertefelt, Lieve De Wachter and Serge Verlinde
Leuven Language Institute, KU Leuven, Dekenstraat 6, 3000 Leuven, Belgium
Keywords: Electronic Writing Aid, Higher Education Students, Writing Skills Development.
Abstract: Students at universities and colleges in Belgium and abroad often experience difficulties with writing
(academic) texts in their native language (De Wachter and Heeren, 2011; Dugan and Polanski, 2006; Gray
et al., 2005; Napolitano and Stent, 2009). This is reflected in many initiatives that are being developed
specifically to support students’ writing skills, among other the development of electronic writing assistance
systems. Many of these systems are based on Natural Language Processing techniques, such as parsing. In
this paper, we will argue that writing aids do not always have to make use of NLP techniques in order to
analyze texts in a detailed and accurate way. We present an online writing aid, Writing Aid Dutch, which
marks possible areas of concern in students’ texts on three levels: (1) text structure and cohesion, (2) style
and (3) spelling and provides users with individualized feedback. Writing Aid Dutch uses a lot of data and
analyzes texts using complex queries and string matching techniques. Initial user experiences have been
very positive so far. From February 2014 onwards, the effectiveness of the writing aid will be investigated
in a one-group pre-post test design.
Students at Flemish universities and colleges often
have difficulties with writing, irrespective of the
educational field they are in (Berckmoes and
Rombouts, 2009; Berckmoes et al., 2010; Bonset,
2010; De Vries and Van der Westen, 2008; De
Wachter and Heeren, 2011; Peters and Van Houtven,
2010). In 2011, a quantitative and qualitative needs
analysis carried out among first year students of KU
Leuven (Belgium) revealed that the most frequent
writing problems of students are situated on the level
of (1) text structure and cohesion, (2) style and, to a
lesser extent, (3) spelling (De Wachter and Heeren,
2011). The results of this needs analysis are
strikingly similar to those of previously conducted
studies in Flanders as well as abroad.
The concern of students’ poor writing skills is
not confined to Belgium alone but is shared
internationally and has already resulted in many
initiatives offering writing support for students
(Taylor and Paine, 1993; Gray et al., 2005; Dugan
and Polanski, 2006; Graham and Perin, 2007).
Among other things is the development of automatic
and semi-automatic writing aids. Desktop
applications such as SWAN (Scientific Writing
AssistaNt, Kinnunen et al., 2012) or web
applications such as the Language Tool Style and
Grammar Checker (Naber, 2014) or Spell Check
Plus (Nadashi and Sinclair, 2014) offer writing
assistance to students who write at an L2 level or in
their native language. These tools often use NLP
techniques, such as a parser, to analyze the inserted
texts in a detailed way.
Many of the writing assistance systems available
today are able to provide students with useful and
accurate feedback on different aspects of their text.
However, despite the good intentions that they have,
some of these writing assistance systems have some
drawbacks as well. In the first place, the accuracy of
the suggested feedback or corrections is not always
satisfactory. Secondly, some of these writing aids,
such as Scientific Writing AssistaNt, are rather time-
consuming as students have to pass several ‘stages’
before receiving any feedback on their text.
Moreover, SWAN provides the user with an
overwhelming amount of information, which makes
that he loses sight of the relevant feedback. This
reduces the feeling of being responsible for your
own writing product as well. Contrary to that, many
D’Hertefelt M., De Wachter L. and Verlinde S..
Writing Aid Dutch - Supporting Students’ Writing Skills by Means of a String and Pattern Matching based Web Application.
DOI: 10.5220/0004948704860491
In Proceedings of the 6th International Conference on Computer Supported Education (CSEDU-2014), pages 486-491
ISBN: 978-989-758-020-8
2014 SCITEPRESS (Science and Technology Publications, Lda.)
web-based writing aids provide too limited
feedback, which leaves the user frustrated and
unsatisfied. Lastly, many writing aids concentrate
too little on the writing process and do not
encourage students’ writing skills development,
because they immediately suggest corrections
(Napolitano and Stent, 2009).
In this paper, we present an online writing aid,
the Writing Aid Dutch, a web application that
responds to the strong need for effective writing
support in Dutch. The writing aid analyzes texts,
using string and pattern matching techniques to
identify errors but also possible areas of concern in
the submitted text. Based on the results of several
needs analyses, the didactic purpose of the writing
aid is to raise students’ awareness on frequent
writing problems that are situated on the level of (1)
text structure and cohesion, (2) style and (3) spelling
(Berckmoes et al., 2010; De Wachter and Heeren,
2011; Peters and Van Houtven, 2010). The writing
aid does not correct and ‘judge’ students’ writing
mistakes, but marks them in the text and provides
students with concise feedback, tips, examples and
links to informative websites. Students can submit
different genres of texts into the writing aid, such as
a report, paper, essay, articles or master thesis.
In what follows, we will discuss the design and
metrics of the writing aid after a short section on
related work. We will then report some first user
experiences and discuss future work, before we turn
to our conclusions.
The development of Writing Aid Dutch fits in with
an international trend of responding to students’
writing problems with the development of electronic
writing assistance systems. More specifically, it
corresponds to the attention shift from product
assessment to process-oriented support (Dale and
Kilgarriff, 2011; Fontana et al, 2006; Gikandi et al.,
2011). Writing assistance systems such as Amadeus
(Fontana et al., 2006) or Helping Our Own (Dale
and Kilgarriff, 2011) are specifically being
developed to assist students throughout their writing
The underlying NLP techniques that these
writing assistance systems use, however, differ from
the data and string and pattern matching techniques
that are implemented in Writing Aid Dutch. Apart
from SOS-Frans (“SOS French”) (Rymenams et al.,
2012), a writing aid aimed at non-native speakers of
French that has been developed at the same institute
as Writing Aid Dutch, there is no knowledge of
writing aids that do not make use of NLP
3.1 Interface
The interface of Writing Aid Dutch is simple and
user-friendly: after students have copy-pasted or
keyed in their text in the input field, they can click
on three coloured buttons that each represent one of
the three problem areas: (1) text structure and
cohesion, (2) style and (3) spelling. These buttons
are connected with arrows indicating the preferred
order in which students should check the text.
However, the student remains free to click on the
button they prefer. As such, a learning path is
suggested but students are free to determine their
own pace in that they can choose which analyzed
elements they want to look at first and when they
want to take another step. The environment of
Writing Aid Dutch is strongly user-controlled,
seeing that our students are rather advanced learners
and therefore do not need maximal guidance.
Moreover, a system that is fully program-controlled
would reduce the motivation of our students.
Figure 1: The three buttons ‘Structure and cohesion’,
‘Style’ and ‘Spelling’ on which students can click.
Considering that the writing tool is being developed
for Dutch native speakers, feedback is in the form of
general advice that is deliberately kept concise in
order not to reduce students’ motivation. For some
of the text elements marked in the text, additional
information is given in small pop-up screens that
appear when the user scrolls over a highlighted text
element, or in an extra field when students click on
‘read more’. The illustration below gives a
screenshot of text analysis and feedback for the use
of structure words. When the user scrolls over a
marked structure word, its meaning is provided in an
extra pop-up field: in the illustration below, the
meaning tegenstelling “contrast” is given for the
structure word echter “however”.
Figure 2: Marking of structure words under ‘Structure and
3.2 Metrics and Implementation
In each of the three levels, students can check
specific textual elements or metrics that are related
to it. In the following sections, the individual metrics
of each level and the data involved will be
3.2.1 Level 1: Text Structure and Cohesion
In the level of text structure and cohesion the student
can check (1) use of reference words, (2) use of
structure words, (3) most frequent words of the text,
(4) recurring sentence patterns, (5) sentence length
and (6) paragraph length. More general statistics
concerning text structure and cohesion, viz. the total
number of words, sentences and paragraphs of the
text are given as well. Lastly, the readability index
(or complexity index) of the text is calculated.
Reference words and structure words are
highlighted in the text by matching the text with lists
of words. For the third metric, namely that of the
most frequent content words of the text, the text is
matched with a frequency list containing word forms
of only content words. The word forms that are
found in the text are lemmatized, and these lemmas
are displayed to the student. As far as the next metric
of recurring sentence patterns concerns, there is no
specific measure. We have worked as follows:
sentences that start with de “the”, het “the”, een “a”,
die “those”, dat “that”, deze “these”, dit “this”, men
“one”, er “there” point out to few variation in
sentence construction. If more than two sentences in
five start with these words, they are marked. This
formula applies to other recurring words as well. For
the last two metrics, sentence and paragraph length,
a minimal and maximal boundary is set: sentences
containing less than 8 words and more than 30
words are marked; the boundaries of the paragraphs
are set at respectively 4 and 17 sentences per
paragraph. For these two metrics, the average
sentence and paragraph length is calculated and
visualized through a small traffic sign, displaying a
red (“too long/short sentences/paragraphs”), orange
or (“possibly too long/short sentences/paragraphs)
green (“sentence and paragraph length confirms to
norm”) light.
The readability index that is calculated is partly
based on the Flesch-Douma formula, the readability
formula based on Flesch (1948) but adapted to
Dutch, which predicts a text’s readability by taking
into account word length, i.e. the number of
syllables per word, and sentence length, i.e. the
number of words per sentence. Despite a number of
objections, such as the idea that long sentences are
not always more complex than shorter ones (Jansen
and Lentz, 2008), this formula has proven to be a
reliable predictor of a text’s readability and
complexity. However, to make the formula even
more accurate we have added word frequency,
seeing that words that are highly frequent are more
understandable than infrequent words. We use a
frequency list consisting of word forms instead of
3.2.2 Level 2: Style
The metrics distinguished in the second level are (1)
use of passives, (3) use of nominalizations, (3)
personal language use, (4) long-winded
constructions, (5) informal and subjective words, (6)
formal and archaic words, (7) vague words and (8)
word combinations. For each of these metrics,
Writing Aid Dutch checks whether the style of the
inserted text is adapted to the required norm. Seeing
that the students who use the writing aid come from
different institutions (university or college) and, as a
consequence, write in different text genres, the
writing aid does not ‘judge’ the inserted text but
provides the student with nuanced information about
these different style requirements. Again, most of
the metrics in this level are highlighted in the text by
string and pattern matching.
3.2.3 Level 3: Spelling
The last level on which students can check their text
is spelling, where typing mistakes and wrongly
spelled words are marked by a spell-checker. The
use of abbreviations is checked as well.
The implementation of the spell-checker has
been (and still is) a labour-intensive work. The spell-
checker is based on a word list containing over
seven hundred thousand words forms that is still
being completed. The database word list contains
headwords supplied with linguistic information such
as word class, article, plural form, past form,
participle etc. In total, fifteen word classes are
The spell-checker functions in various steps. The
process starts by distinguishing every word
separately, defining its boundaries by marking the
spaces and punctuation marks and as such splitting
up the sentence. After sentences are subdivided into
separate words, occurrences of more or less fixed
expressions are first of all being looked at. The
database contains a list of these expressions,
especially archaic phrases, which is matched with
the text. A second step checks whether the
remaining unrecognized and single words are in the
word list. When this is not the case, the word will
have to pass several conditions before it will be
marked as wrong. In what follows, we will describe
some of these conditions.
A first condition comprises combinations of
numbers followed by a special character that are
allowed in academic papers, for example “5°” or
“10%”. A second condition refers to other symbols
that may occur as well, such as Roman numbers like
“I”, “IV” or “XI”. For the third and the fourth
criterion, it is important to note that Dutch is a
compound language in which words can very easily
be composited. Compounds in Dutch are always
written in one word or with a hyphen. The third
selection criterion then concerns compound words
that are written with a hyphen and consist of words
that also exist on their own, for example a word such
as adjunct-directeur “adjunct-director”. The fourth
condition picks out compounds written without a
hyphen. In this step, two functions are used to
reduce the number of possibilities. A first one splits
up a word, for example the word strooizout “road
salt”, in the following manner:
The function stops when both queries give a
valuable result, in this case strooi and zout. The
minimal length for a word to be recognized is fixed
at four characters, seeing that fewer characters
resulted in too many false positives, i.e. words that
do not exist but are nonetheless grammatically
correct. A second function in this condition relates to
the syntactic place that a particular word can have in
a compound, namely in the beginning or at the end
of the compound. This is statistically determined on
the basis of the word list. For each syntactic option,
frequency is calculated. For example, achterover
“back” can never occur at the end of a compound
but occurs, so far, a hundred and nine times in the
beginning of a compound word, like in the verb
achteroverleunen “to lean back”. In the fifth step of
process, the spell-checker looks at a list containing
named entities. When a word, then, still has not been
found, the context is taken into account in order to
check whether the word is part of a word group that
has not been recognized as a fixed expression.
Concretely, the context is limited to a span of four
words left and right.
When a word still has not been recognized after
these selection criteria, it will be marked red in the
students’ text. However, a word can also be marked
blue in the text. For these words, the spell-checker
suggests an alternative form, based on the
Levenshtein distance principle. This principle tries
to alter one string into another string by making
minimal changes, for example by changing or
deleting one letter. The spell-checker is designed in
a way that it is partly self-supportive. Unrecognized
words automatically appear in a separate database,
so that they, in the case of correct words, may be
added later to the spell-check word list.
3.3 Comparison to Word Processing
Software Such as Microsoft Word©
In Microsoft Word© grammar and spelling can be
checked in a variety of languages, among which is
also Dutch. A comparison between Microsoft
Word© and Writing Aid Dutch seems therefore
relevant. With regard to the computational
implementation, language-specific information in
Writing Aid Dutch cannot, unlike in Microsoft
Word©, be considered as a rule set that is imported
in the system. In the spell-checker of the writing aid,
for example, many of the hard codes are only
applicable to Dutch. An example is the following
part of a code:
The part alleen in samenstelling “only in compound”
relates to complex verbs in Dutch such as
tekeergaan “to rant”. The part tekeer does not exist
on its own but always occurs in combination with
the verb gaan “to go”; as a consequence, tekeer will
not be marked wrong because it is part of a complex
verb. However, the codes that are used in Writing
Aid Dutch to refer to its underlying databases can
easily be adapted to other languages; only the
databases itself will be different.
Because of the many complex and language-
specific codes, the spell checker of Writing Aid
Dutch is much more accurate and complete than the
Dutch spell checker in Microsoft Word©. Checking
grammar has never been a priority in the
development of Writing Aid Dutch, seeing that its
target audience are advanced native speakers of
4.1 Text Analysis on Content Level
At the moment, we are also experimenting with
more content-oriented text analysis by categorizing
certain words that appear in a student’s text into
semantic fields. For this experiment we have used
texts of KU Leuven students of Political Science, in
which they had to compare two politicians. By
identifying these words that express either similarity
or difference in the text, the distribution of these two
semantic categories is revealed, so that it can be
investigated if they appear equally and at the right
place in the text. Another experiment is the
identification of academic words or more technical
terminology in the text.
4.2 Effectiveness Analysis and Further
User Study
From February 2014 onwards we will investigate the
effectiveness of the writing aid in a quantitative and
qualitative one-group design study. Despite the fact
that such a design has minimal internal validity and
no external validity (Sytsma, 2002), we have chosen
this design because of time restrictions of the
project. A within-subjects design does not require a
placement test that cancels out possible differences
in competencies between participants (de Smet et al.,
2011). A total number of minimal 60 students of
university as well as college institutions will be
tested. On the one hand, effectiveness will be
measured by rating texts written without and written
with Writing Aid Dutch. On the other hand,
students’ as well as teachers’ perception of the
learning progress will be evaluated. The results of
the effectiveness experiment will be available in
June 2014.
A tool that is similar to Writing Aid Dutch, SOS-
Frans, has been developed at the KU Leuven for
French as a second and foreign language and turned
out to be very effective, leading to fewer mistakes
(Rymenams et al., 2012). Scientific Writing
AssistaNt, reduced the lack of structure and
semantic coherence in scientific papers (Kinnunen et
al., 2012). Moreover, as teachers, we have already
experienced noticeable progress in papers of
students when they use Writing Aid Dutch. By
analogy with similar writing aids and on the basis of
our experiences, we hypothesize that the learning-
process of students who use the Writing Aid Dutch
will improve and that their writing products will be
As mentioned in Leakey (2011), the empirical
data that result from quantitative research should
ideally be completed with judgmental data. We have
already gathered initial user experience by means of
an online questionnaire filled in by 50 students. Next
to students, 10 teachers of several faculties have
reported their experiences in focus interviews.
However, these data are not sufficient and we will
carry out extra questionnaires and focus interviews
with students and teachers as part of our
effectiveness study.
In this paper, we have presented the Writing Aid
Dutch. We have shown that the implementation of
NLP techniques is not always a prerequisite for the
development of appropriate computer-based support.
Text analysis based on string and pattern matching
techniques can be detailed, correct and fast. The
writing aid (1) raises students’ awareness of frequent
writing issues, (2) provides clear and individualized
feedback, tips and examples, (3) focuses on the
process, (4) has a simple and user-friendly design
and (5) leads to less ‘shallow’ and repetitive
correction work for lecturers. As a web application,
the writing aid is a durable and partly self-supportive
tool that can be adapted at any time.
Berckmoes, D., Rombouts, H., 2009. Rapport verkennend
onderzoek naar knelpunten taalvaardigheid in het
hoger onderwijs. [Report preliminary investigation of
higher education students’ difficulties of literacy
skills]. Antwerp: Linguapolis/University of Antwerp.
Available at: <http://webh01.ua.ac.be/linguapolis/
Monitoraat_op_maat.pdf> (Accessed 27 June 2012).
Berckmoes, D., Rombouts, H., Hertogs, K., 2010.
Taalstimulering academisch Nederlands voor
studenten aan de Universiteit Antwerpen. Monitoraat
op maat. Rapport derde jaar, september 2008 –
augustus 2009 (Language stimulation academic Dutch
for students at the University of Antwerp. Report third
year). Linguapolis/University of Antwerp.
Bonset, H., 2010. Nederlands in voortgezet en hoger
onderwijs: hoe sluit dat aan? (Dutch in secondary and
higher education: how does it match?). Levende talen
magazine, 97(3), pp. 16-20.
Dale, R., Kilgarriff, A. 2011. Helping Our Own: The HOO
2011 Pilot Shared Task. Proceedings of the 13th
European Workshop on Natural Language
Generation, pp. 242–249.
De Smet, M. J. R., Broekkamp, H., Brand-Gruwel, S.,
Kirschner, P. A., 2011. Effects of electronic outlining
on students’ argumentative writing performance.
Journal of computer assisted learning, 27, pp. 557-
De Vries, H., Van der Westen, W., 2008. Talige
competentie in het hoger onderwijs (Linguistic
competence in higher education). In: S. Vanhooren, A.
Mottard, eds. 2008. 22ste conferentie Het Schoolvak
Nederlands. Gent: Academia Press. pp. 115-120.
De Wachter, L., Heeren, J., 2011. Taalvaardig aan de start.
Een behoefteanalyse rond taalproblemen en
remediëring van eerstejaarsstudenten aan de
KULeuven (Entry-level academic language skills. A
needs analysis of language problems and remedy of
first year university students at the University of
Leuven). Leuven Language Institute/University of
Leuven. Retrieved from https://ilt.kuleuven.be/
Dugan, R. F. Jr., Polanski, V. G., 2006. Writing for
computer science: a taxonomy of writing tasks and
general advice. Journal of Computing Sciences in
Colleges, 21(6), pp. 191-203.
Flesch. R., 1948. A new readability yardstick. Journal of
Applied Psychology, 32, pp. 221-233.
Fontana, N. M., Caldeira, S. M. A., De Oliveira, L. C. F.,
Oliveira Jr., O.N. 2006. Computer assisted writing.
Applications to English as a foreign language. CALL,
6(2), pp. 145-161.
Gikandi, J. W., Morrow, D., Davis, N. E. 2011. Online
formative assessment in higher education: a review of
the literature. Computers and Education, 57, pp. 2333-
Graham, S., Perin, D., 2007. Writing next: Effective
strategies to improve writing of adolescents in middle
and high schools – A report to Carnegie Corporation
of New York. Washington, DC: Alliance for Excellent
Education. Available at: http://www.all4ed.org/
files/WritingNext.pdf (Accessed 12 September 2012).
Gray, E. F., Emerson, L., MacKay, B., 2005. Meeting the
demands of the workplace: science students and
written skills. Journal of science education and
technology, 14(4), pp. 425-435.
Jansen, C., Lentz, L., 2008. Hoe begrijpelijk is mijn tekst?
De opkomst, neergang en terugkeer van
leesbaarheidsformules (How understandable is my
text? The rise, downfall and comeback of readability
formulas). Available at: http://www.kennislink.nl/
publicaties/hoe-begrijpelijk-is-mijn-tekst (Accessed 1
January 2014).
Kinnunen, T., Leisma, H., Machunik, M., Kakkonen, T.,
Lebrun, J. L., 2012. SWAN – Scientific Writing
AssistaNt. A tool for helping scholars to write reader-
friendly manuscripts. Proceedings of the 13th
conference of the European chapter of the association
for computational linguistics, pp. 20-24.
Leakey, J., 2011. Evaluating computer-assisted language
learning. An integrated approach to effectiveness
research in CALL. Bern: Peter Lang.
Naber, D. 2014. Language Tool Style and Grammar
Checker. Available at: www.languagetool.org
(Accessed 1 January 2014).
Nadashi, T., Sinclair, S., 2001-2014. Spell Check Plus.
Nadaclair Language Technologies. Available at: <
http://spellcheckplus.com> (Accessed 1 January
Napolitano, D. M., Stent, A., 2009. TechWriter: an
evolving system for writing assistance for advanced
learners of English. CALICO Journal, 26(3), pp. 611-
Peters, E., Van Houtven, T., 2010. De weg naar
materiaalontwikkeling is geplaveid met behoeftes (The
way to material design is paved with needs). In: E.
Peters, T. Van Houtven, eds. 2010. Taalbeleid in het
hoger onderwijs. De hype voorbij?. Leuven: Acco. pp.
Rymenams, S., Verlinde, S., Marx, S., Rosselle, M.,
Gerard, L., 2012. SOS-français: conception et
évaluation d’un didacticiel d’aide à la rédaction
interactif (SOS-French: development and evaluation of
a didactic aid in an interactive environment). 3e
Congrès Mondial de Linguistique Française. SHS Web
of Conferences, 1, pp. 377-393.
Sytsma, S., 2002. The basics of experimental design.
Available at: http://courses.washington.edu/bio480/
Basics_of_Experimental_Design.pdf (Accesses 20
November 2012).
Taylor, H. G., Paine, K. M., 1993. An inter-disciplinary
approach to the development of writing skills in
computer science students. Proceedings of the
SIGCSE Technical Symposium on Computer Science
Education, pp. 274-278.