An Authoring Tool based on Semi-automatic Generators for Creating
Self-assessment Exercises
Nathalie Guin
a
and Marie Lefevre
b
Univ. Lyon, UCBL, CNRS, INSA Lyon, LIRIS, UMR5205, F-69622 Villeurbanne, France
Keywords: Self-assessment, Semi-automatic Generation of Exercises, Authoring Tool, University Context.
Abstract: This article presents ASKER, a tool for teachers to create and disseminate self-assessment exercises for their
students. Currently used in the first year of a bachelor's degree at the University of Lyon (France), it enables
students to carry out exercises in order to evaluate their acquisition of concepts considered important by the
teacher. ASKER enables the creation of exercises (matching, grouping, short open-ended questions, multiple
choice questions) that can be used to assess learning in many different fields. To create exercises to assess a
concept, the teacher defines a model of exercises that will enable the generation of various exercises, using
text or image resources. Such an exercise model is based on constraints that the exercises created from this
model must comply with. Automatic generators create, from the resources defined by the teacher, many
exercises respecting these constraints. The possibility for the learner to request the generation of several
exercises from the same model enables her to assess herself several times on the same concept, without the
teacher having to repeatedly define many exercises.
1 INTRODUCTION
The purpose of the work described in this article is to
enable a learner to assess her knowledge within a
learning path for which a teacher defines learning
objectives. Our approach is to enable the teacher to
provide the learner with exercises on the concepts to
be acquired. The learner can use these exercises if she
wishes to evaluate her mastery of the concepts
involved in the exercises. We therefore place
ourselves here in the context of a formative
evaluation.
Self-assessment with immediate feedback
requires activities or features that allow students to
assess themselves against the course objectives.
Exercisers (or exercise generators) are a way to
quickly diagnose the skills acquired, to perform
performance memorization and skill development
through trial and error learning based on repetition
(Mostow et al., 2004). Since the exerciser
environment keeps track of the learning activity and
provides immediate feedback, it facilitates the
learner’s regulation by allowing explicit reflection on
the skills worked on (Steffens, 2006).
a
https://orcid.org/0000-0001-9999-9878
b
https://orcid.org/0000-0002-2360-8727
We propose an authoring tool that gives the teacher
the freedom to set the notions on which learners can
assess themselves, and that enables him or her to
create exercises to assess the mastery of these notions.
In order to meet the needs of as many people as
possible, we have chosen to consider types of
exercises that can be applied to many fields, such as
MCQ (Multipe Choice Questions), matching,
grouping, ordering, gap texts and so on.
The following section explains why we have
chosen an approach based on exercise generators to
address this issue of creating self-assessment
exercises. We then specify the knowledge acquisition
processes necessary for this approach, before
presenting the architecture of the tool we have
developed: ASKER (Authoring tool for aSsessing
Knowledge genErating exeRcises). Finally, after
having shown how this tool can be used by both
teachers and learners, we carry out an evaluation of
the use of ASKER in first-year bachelor’s degree
courses at the University of Lyon.
Guin, N. and Lefevre, M.
An Authoring Tool based on Semi-automatic Generators for Creating Self-assessment Exercises.
DOI: 10.5220/0010996100003182
In Proceedings of the 14th International Conference on Computer Supported Education (CSEDU 2022) - Volume 1, pages 181-188
ISBN: 978-989-758-562-3; ISSN: 2184-5026
Copyright
c
2022 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
181
2 AN AUTHORING TOOL BASED
ON EXERCISE GENERATORS
Our aim was to enable a teacher to offer learners
online self-assessment exercises. Such exercises
enable learners to autonomously test their level of
mastery of what they have learned in the course,
whether this course is online or face-to-face. Learners
may fail in initial attempts to respond to exercises if
knowledge is not mastered. It is therefore possible
that a learner may have to answer several times to the
same exercise before achieving success. In order for
the learner not to be influenced by her previous
resolutions, it is necessary that the self-assessment
exercises be different from one time to another, while
evaluating the same knowledge. However, it does not
seem reasonable to ask the author to write many
versions of the exercise. That's why we have chosen
to use exercise generators that the author can easily
use in any field. However, we designed a semi-
automatic process of generating exercises, in order to
let teachers take part in the choice of the criteria that
the exercises will have to meet.
To meet the needs of teachers and learners, the
expected properties of the authoring tool were as
follows:
- The exercise is different from one time to another.
- The author is in control of the content of the exercise
and is assured that it meets his or her expectations
in terms of educational content.
- Exercise generators can be used in a wide range of
fields and grade levels.
- The diagnosis of the response is made automatically
and in real time.
- The construction of exercises with generators is a
time saving for the author compared to a creation
exercise by exercise.
- The creation of exercises does not require technical
skills.
Many researches have studied the question of
authoring tools in the field of Technology Enhanced
Learning, and several literature reviews about the
topic were published (Murray, 1999) (Murray, 2003)
(Woolf, 2010) (Dermeval et al., 2016). In these
works, the objective is to create Intelligent Tutoring
Systems. For our part, we only want to enable the
creation of training and self-assessment exercises,
which is why we focus our study on exercise
generators. As we also want our authoring tool to be
usable in any field, we have not integrated a model of
1
https://articulate.com/360/studio#quizmaker
2
http://hotpot.uvic.ca/
3
https://www.claroline.com/
the knowledge and skills to be assessed. Indeed, our
goal is that a teacher can directly use the tool to create
exercises without the need for a prior knowledge
modeling phase. We thus agree with Baker's approach
(Baker, 2016), which, noting that ITS used at scale
are not the most intelligent ones, proposes to design
simpler tools that support teachers' activity by
enhancing their expertise.
Existing exercise generators can be categorized
into three categories: manual, automatic and semi-
automatic.
Often used as part of authoring tools, manual
generators give a great deal of freedom to the author,
which precisely defines the content of the exercise
and all the formatting options. Some commercially
available authoring tools such as Articulate
Quizmaker
1
or Hot Potatoes
2
are commonly used to
create paper-based or computer-based exercises. The
online learning platforms Claroline
3
and Moodle
4
,
commonly used in higher education, also offer their
own exercise editing tools. With this type of
generators, the author is guaranteed to have an
exercise that precisely matches his/her expectations,
which meets one of our needs. On the other hand, the
author must create each instance of exercise one by
one, specifying its contents. This type of generator is
not able to automatically create a large number of
exercises assessing the same skill.
Automatic generators do have this ability, but
unfortunately leave little space for the author in the
creation process. With this kind of generator, a large
number of exercises are created automatically
without the author being able to influence the
system's choices. He or she can simply choose the
category of the exercise (form, theme, knowledge
addressed) but cannot act on the content or on specific
constraints. Examples include the Reading Tutor
generator (Mostow et al., 2004), or the Aplusix
generator (Bouhineau et al., 2008).
In order to take advantage of the features of the
automatic generators while leaving to the author the
editorial freedom on the exercises created, we have
chosen semi-automatic generators, which combine
the advantages of the two previous categories. These
generators propose that the author define a model of
exercises, which is then instantiated to produce a
large number of exercises (Jean-Daubias and Guin,
2009) (Delozanne et al., 2003). Some e-learning
platforms like Moodle, Wims
5
, WeBWorK
6
or
(Auzende et al., 2007) offer exercises involving
4
http://moodle.org
5
https://wims.univ-cotedazur.fr/wims/
6
http://webwork.maa.org/
CSEDU 2022 - 14th International Conference on Computer Supported Education
182
variables that are similar to the concept of an exercise
model. This type of generator partially meets our
needs but is limited to the areas requiring calculation.
They are used for fields such as mathematics and
science, and not all of them are accessible to non-
programmer authors.
In order to have semi-automatic generators
adapted to many domains and including other types
of exercises, we have chosen to use the GEPPETO
approach (Lefevre et al., 2012). This approach
consists of enabling the teacher to express constraints
on the exercises to be generated. To do this, it is
necessary to have a model of the types of exercises
that can be generated, in order to know the type of
constraints that the teacher must be able to express.
The following section thus specifies the models
guiding the acquisition of the knowledge necessary to
generate exercises according to the GEPPETO
approach.
3 ACQUISITION OF
KNOWLEDGE FOR
GENERATING EXERCISES
Figure 1 presents the GEPPETO approach (GEneric
models and processes to Personalize learners'
PEdagogical activities according to Teaching
Objectives). This approach enables the acquisition of
knowledge at several levels, from experts and
teachers, to generate exercises.
In GEPPETO, a meta-meta-model of exercises
was defined by the research team (Lefevre et al.,
2009). This model specifies the knowledge that an
expert will need to define in order to create a meta-
model of exercises of a given type (see A in Figure
1), for example a meta-model of exercises of the
MCQ type, or of the matching type.
This meta-model of exercises of type X or Y then
enables the teacher to specify constraints defining a
model of exercises (cf. B in Figure 1). Depending on
the type of exercises, the constraints will not be the
same, so that’s why the exercise meta-model is
needed. For example, the teacher could use the meta-
model of the MCQ-type exercises to define a model
of MCQ-type exercises covering a given subject and
containing N questions, with for each M propositions
with only one correct answer.
Using such exercise models, exercise generators
can construct several exercises conforming to these
models (see C in Figure 1). The exercise generators
able to use these models depend on the type of
exercises, and therefore on the meta-model of
exercises of type X or Y.
It can therefore be seen that the GEPPETO
approach requires two knowledge acquisition
processes:
- Acquiring the expert's knowledge for the creation of
meta-models of exercises (see A in Figure 1). This
acquisition is done only once for each type of
exercises and is based on the meta-meta-model of
exercises.
- The acquisition of the teacher's knowledge for the
creation of the exercises models (see B in Figure
1). This acquisition is carried out on a regular
basis for the construction of various exercise
models and requires an interface that is based on
the meta-model of exercises of a given type (the
constraints that the teacher must define depend on
the type of exercises).
Figure 1: The GEPPETO approach.
An Authoring Tool based on Semi-automatic Generators for Creating Self-assessment Exercises
183
Meta-models and the meta-meta-model are
independent of the field for which an exercise will be
generated. For example, the meta-model
Identification of text parts specifies how to formulate
the guidelines, how to characterize the different text-
based resources, and how to describe the actions to be
carried out on these resources to generate exercises.
An exercise generator is associated to each of
these meta-models. An interface associated with the
generator and based on the meta-model enables the
teacher to define constraints on the exercises to be
generated. It is at this point, when creating the model
of exercises (for example, a gap text where it is
necessary to put the verbs in the past), that the
application to a field and a level of study will be
carried out. The exercises generated from the exercise
model are therefore, of course, dependent on the field.
Since the meta-models are all consistent with the
meta-meta-model, all the exercise generators share a
common architecture (Lefevre et al., 2009).
4 ASKER TOOL
ARCHITECTURE
The GEPPETO approach was designed to create
paper-pencil exercises. To design an authoring tool to
provide learners with online exercises and immediate
diagnosis, we chose to follow the same approach. Our
research hypothesis was that using constraints on the
exercises to be generated allowed both to obtain a
sufficient variety of exercises for learners to train and
self-assess, while requiring less work for the author
teacher.
Thus, in the ASKER authoring tool, we have
chosen a model inspired by the GEPPETO approach:
after having chosen a type of exercise (MCQ,
matching, gap text...), the author can create an
exercise model that describes the content and form of
the exercises he or she wants to create, but without
necessarily constraining it completely. By exploiting
this model, an exercise generator is able to provide
the learner with a large number of different exercises
evaluating the same skill. Each exercise instance
generated in this way will be interactive: the learner
will respond online and obtain a diagnosis of her
response.
The types of exercises we selected are:
identification of parts of the text (includes gap text),
organization of elements (ordering, grouping,
matching), annotation of illustrations, Multiple
Choice Questionnaire, open and short-ended
questionnaire.
Figure 2 shows the architecture of the ASKER
authoring tool. The upper block is made up of the
different levels of representation of the exercises,
resulting from the GEPPETO approach: the meta-
models of types of exercises, the models of exercises,
and the instances of exercises. In the central block are
the three mechanisms that manipulate these
representations of exercises. The lower block
contains the resources used in the exercise creation
process.
Resources are the basic elements that are used to
build exercises, for example texts, images or element
sequences. Each resource has metadata characterizing
it to facilitate searches (theme, level, etc.) as well as
metadata enriching the resource to define exercises,
such as image captions or annotations on image areas.
For example, the author can define image type
resources, with flag images, and define for each flag
image its country, its capital city, and its continent.
Figure 2: ASKER tool architecture.
CSEDU 2022 - 14th International Conference on Computer Supported Education
184
The author chooses a type of exercises T (for example
matching). Let’s call MMT the meta-model of the T-
type exercises described by an expert. The author
creates a model of exercises (let’s call it MT)
compliant with MMT using a dedicated interface
based on the knowledge contained in MMT about the
type of exercises T. This MT model defines a set of
constraints that the exercises resulting from this
model must respect. For example, it is a 5 pair
matching exercise, choosing flag images, and
matching each image with its "country" metadata.
The author can generate some instances exercises of
MT (let’s call them ExoT) to test if the model does
give rise to the expected exercises
The T-type exercises generator associated with
MMT (so here the matching exercises generator)
therefore receives an input model of exercises MT
that it instantiates to produce an output ExoT
exercise. The ExoT exercise is consistent with the MT
model and therefore with the choices of the author
who created it. The generator requires no human
intervention. It has all the necessary information in
the MT exercise model and makes use of resources
(here the flag images). The generator is used
whenever you want to obtain a new ExoT instance of
exercise (which contains the diagnosis) from the MT
model.
The exercise is then presented to the learner via a
resolution tool that formats the exercise, gathers the
learner's answer and provides a diagnosis. In our
example, the generated exercise will propose 5 flags
to the student, and the student will have to find the
country of each flag. The variety of exercises
generated will therefore come in this example from
the amount of flag images available in the resources.
The same applies to exercises using texts. Variety can
also come from the use of formulas using variables
whose values must be chosen within an interval
defined by the author. The resources can be used for
different exercise models. For example, we could
define an exercise model where the capital associated
with each flag must be found. Or a categorization
exercise where flags must be put in boxes
corresponding to their continent.
5 THE ASKER TOOL
The ASKER (Authoring tool for aSsessing
Knowledge genErating exeRcises) platform can be
used on the one hand by a teacher to create models of
exercises, and on the other hand by learners to carry
out exercises generated from these models and to
obtain a diagnosis for self-assessment.
An Authoring Tool for the Teacher. ASKER
currently enables a teacher to create models of
exercises of type MCQ, short open-ended questions,
matching, grouping and ordering. The teacher can
create resources such as texts, images, MCQ
questions, short open-ended questions. On each of
these resources, he or she can add meta-data that will
be used by the exercise models using the resource. A
resource can be used for several exercise models of
different types. Thus, the same meta-data on a
resource can be used for both a matching, grouping or
ordering exercises.
Figure 3: Author’s interface for creating a model of matching exercises.
An Authoring Tool based on Semi-automatic Generators for Creating Self-assessment Exercises
185
Figure 3 shows the author's interface for creating a
model of matching exercises. The author selected
text-based resources (right-hand side of Figure 3) and
filtered out those for CC (chapter) number 3 of UE
LIF3 (the name of the teaching unit). Several texts of
functions in Scheme programming language can be
used. On these texts, meta-data specifies the
specification of each function, its input type(s) and its
output type.
To create an exercise model, the author defines
(left part of Figure 3) that he or she wishes to create
exercises in which 4 functions must be associated
with their specification. The exercises generator then
uses this exercise model to create exercises that meet
these constraints, using the available resources
describing functions. Another model of exercises
could use these same resources describing functions
but asking to group them by input type or output type.
In this way, the same resources can be used to
produce another type of exercise and related to
another notion of the course.
A Self-assessment Tool for the Learner. The
teacher suggests to his or her students models of
exercises corresponding to the concepts studied in
class. For each exercise model, the learner can request
the generation of several instances of exercises. She
then resolves a first exercise derived from the
exercise model, next the system diagnoses her
answers and displays a feedback (in green and red) on
her answers as well as the correct answer (in blue) to
the exercise (see Figure 4).
A commentary prepared by the teacher explaining
a common error or reminding an important concept
may also be displayed. The learner can then revise the
course if necessary and ask for a new exercise based
on the same exercise model.
6 EVALUATION OF THE ASKER
TOOL
At the University of Lyon, in the first year of a
Computer Science degree, there are two initial
courses in programming: one on imperative and
iterative programming in C language, the other on
functional and recursive programming in Scheme
language.
We set up the use of the ASKER tool for the
Scheme programming course several years ago. It is
a use of ASKER for a teaching that is not digitized,
the platform being a complementary tool to face-to-
face teaching. We suggested that the students use
ASKER to self-assess their understanding of the
concepts presented in class before coming to the
supervised works. This enables students to situate
themselves in relation to the acquisition of the notions
covered in courses, to prepare the assessments carried
out each week in supervised works, and to diagnose
their difficulties.
For this purpose, we have proposed a set of model
exercises for each of the 9 lectures. By instantiating
the meta-exercise models, we provided students with
18 models of matching exercises, 9 models of
grouping exercises, and 8 models of MCQs. The
creation of these 35 models and their 121 resources
represented between 1 to 2 hours of work each week,
during the 12 weeks of teaching, for the teacher in
charge of the fall semester. This is a considerable
workload, but it only concerned the initialization
phase. These different models and their resources
were then exploited and completed by the spring
semester teacher in a more reasonable time: 1/2 hour
per week. Since then, the use of ASKER in this course
does not require any more time for the teacher.
Figure 4: Feedback provided by ASKER to the learner on her resolution of several types of exercises.
CSEDU 2022 - 14th International Conference on Computer Supported Education
186
We did not consider evaluating the quality of the
exercises generated in terms of their impact on student
learning. Indeed, our objective being to provide the
teacher with a tool enabling her or him to generate
varied exercises in sufficient number, we consider that
the tool fulfils its objective if the exercises generated
are in accordance with the teachers' expectations,
which is the case with our approach of constraints
defined by the teachers and satisfied by the generators.
We then introduced the use of ASKER in the other
first year course, devoted to algorithms and
programming in C. The teacher has created 37
exercise models, divided into 6 chapters. Of the 584
students enrolled in the course, 485 (83%) used
ASKER at least once. For each model planned by the
teacher, the number of students who did at least one
exercise from this model was around 46%.
In order to measure the impact of the platform on
students, we could not for ethical reasons conduct a
comparative experiment, giving access to ASKER
only to a part of the students. A questionnaire was
distributed at the end of the semester to students
enrolled in the course. We received 106 responses
from students who used the platform as this:
- 67% of students did exercises each week, 23% every
other week, 9% at the beginning of the semester
but no more afterwards.
- At each use, 43% used it between 5 and 10 minutes,
43% between 10 and 20 minutes.
- 82% of students generated multiple instances of
exercises from the same model.
- Students mainly used it to study before the
supervised works sessions (83%). They each time
made all the models of exercises proposed for a
chapter (82%).
- Students reported that ASKER helped them to
understand the course (58%) and to identify
concepts not understood (63%). 89% of them
think that using ASKER has enabled them to
progress in this course (70% a little and 19% a lot).
Although this is not an evidence of ASKER's impact
on learning, students think that this tool has had a
positive impact on their learning. Students are not
under any obligation to use ASKER during their first
year. Their use of the tool is in no way used to evaluate
their work. The tool is just available if they want to
use it and many use it all year round. Usage statistics
and student comments also show that this tool
increases their motivation to work, which in itself is
already a very satisfying result.
ASKER has also been used by 3 physics and
chemistry teachers at high school. These teachers used
to offer their students paper-based exercises to enable
them to self-assess certain skills. They wanted to use
ASKER to produce similar exercises that would give
their students immediate feedback on their answers.
This immediate feedback was a demand for 75% of
their students.
After a while of hands-on learning, these teachers
were able to use ASKER to create 44 exercise models
for their students. They have made extensive use of
image-type resources. They appreciated the
opportunity to have a competency used in various
situations (due to the variety of resources) and to
create exercises involving different cognitive tasks
(due to differing types of exercises). Using ASKER
also gave them the idea of new exercises compared to
those they used on paper. Their students loved the
application, and in particular the immediate feedback.
They have asked to be able to use ASKER in all
chapters of their Physics and Chemistry courses.
To meet the self-assessment requirements that we
formulated in Section 2, our tool had to have the
following properties:
- The exercise is different from one time to another.
This property is satisfied thanks to the generators
using constraints set in the exercises models. The
practice shows that students actually do several
exercises for each exercise model (82% of them).
- The author is in control of the content of the exercise
and is assured that it meets his or her expectations.
This property is satisfied with the authoring tool
that enables the teacher to create an exercise model
specifying the constraints that the exercises must
satisfy, and enable him/her to control the exercises
generated from each model.
- Exercise generators can be used in a wide range of
fields and grade levels. ASKER has been used in
computer science, optics and anatomy at
university, physics and chemistry at high school,
but also to evaluate schoolchildrens' knowledge of
countries around the world or to generate IQ tests
based on logical sequences.
- The diagnosis of the response is made automatically
and in real time. The models of exercises defined
by teachers include the knowledge necessary for
generators to diagnose student responses.
-
The construction of exercises with generators is a
time saving for the author compared to a creation
exercise by exercise. Although the definition of
exercise models takes time, especially at the
beginning, teachers appreciate the variety of
exercises generated by using annotated images
and texts or calculation formulas. Creating such a
variety of exercises without generators would take
too much time.
- The creation of exercises does not require technical
skills. In both programming courses, the teachers
An Authoring Tool based on Semi-automatic Generators for Creating Self-assessment Exercises
187
who have used ASKER are computer scientists,
but the use of ASKER does not require computer
skills. In other uses, the authors were professors of
physics, chemistry or optics. The latter has taken
charge of this tool in complete autonomy.
To conclude, all the feedback from the use of ASKER
in different contexts allows us to consider that the tool
meets the needs of both teachers and learners.
7 CONCLUSION AND
PROSPECTS
In this article we introduced ASKER, a tool that
enables teachers to create self-assessment exercises
for their students. This tool can be used for distance
learning or as a complement to face-to-face teaching.
It enables the creation of exercises (matching,
groupings, short open-ended questions, MCQ) that
can be used to evaluate learning in many fields. To
create exercises to assess a concept, the teacher
defines a model of exercises that will enable the
generation of various exercises, using text or image
resources. The possibility for the learner to request
the generation of several exercises from the same
model enables her to self-assess repeatedly on the
same concepts, without the teacher having to
repeatedly define many exercises.
Our research hypothesis was that using
constraints on the exercises to be generated allowed
both to obtain a sufficient variety of exercises for
learners to train and self-assess, while requiring less
work for the author teacher. The evaluation results,
reported in Section 6, allow us to validate this
research hypothesis.
ASKER is a tool that can be used in a variety of
fields, and in a variety of learning contexts, at any
level. It thus offers many possibilities of use. Its main
limitation is that there is no explicit representation in
ASKER of the knowledge to be learned. The
acquisition of this knowledge therefore represents a
major challenge. The main users of ASKER being the
authors, it would be interesting for them to build the
domain knowledge, as they already do for formulas.
The system could assist them in this task by proposing
a generalization of the information that they provide
to create their models of exercises. We intend to use
activity traces of teachers using ASKER to enable the
system to assist them in this elicitation of domain
knowledge.
We also envisage the use of a particular meta-data
describing the skills mobilized by a model of
exercise, so that we can propose to the learner an open
profile of skills that will enable her to be more
involved in her self-assessment process, for example
by setting objectives to be achieved. Such skills
profiles will also enable us to propose to the student a
learning and training path that will enable her to
achieve such objectives.
REFERENCES
Auzende, O., Giroire, H, Le Calvez, F. (2007). Extension
of IMS-QTI to express constraints on template
variables in mathematics exercises. In 13th
International Conference on Artificial Intelligence in
Education - AIED, Los Angeles, USA, 524-526.
Baker, R.S. (2016). Stupid tutoring systems, intelligent
humans. International Journal of Artificial Intelligence
in Education, 26(2), 600–614.
Bouhineau, D., Chaachoua, H., Nicaud, J.F. (2008).
Helping teachers generate exercises with random
coefficients. Int. journal of continuing engineering
education and life-long learning 18(5-6), 520–5337.
Delozanne, E., Grugeon, B., Prévit, D., Jacoboni, P. (2003).
Supporting teachers when diagnosing their students in
algebra. In Workshop Advanced Technologies for
Mathematics Education Artificial Intelligence in
Education, AIED, Sydney, Australia, 461–470.
Dermeval, D., Paiva, R., Bittencourt, I. et al. (2018).
Authoring Tools for Designing Intelligent Tutoring
Systems: a Systematic Review of the Literature. Int.
Journal of AI in Education 28(3), 336-384.
Jean-Daubias, S., Guin, N. (2009). AMBRE-teacher: a
module helping teachers to generate problems. In 2nd
Workshop on Question Generation, AIED, 43-47.
Lefevre, M., Jean-Daubias, S., Guin, N. (2009). Generation
of pencil and paper exercises to personalize learners'
work sequences: typology of exercises and meta-
architecture for generators. E-Learn 2843-2848.
Lefevre, M., Guin, N., Jean-Daubias, S. (2012). A Generic
Approach for Assisting Teachers During
Personalization of Learners' Activities. Workshop
PALE, UMAP 35-40
Murray, T. (1999). Authoring intelligent tutoring systems:
an analysis of the state of the art. International Journal
of Artificial Intelligence in Education, 10, 98–129.
Murray, T. (2003). An overview of intelligent tutoring
system authoring tools: updated analysis of the state of
the art. In Authoring tools for advanced technology
learning environments, Springer, 491–544.
Mostow, J., Beck, J.-E., Bey, J., Cuneo, A., Sison, J., Tobin,
B., Valeri, J. (2004). Using automated questions to
assess reading comprehension, vocabulary and effects
of tutorial interventions. Technology, Instruction,
Cognition and Learning, Vol. 2, 97–134.
Steffens, K. (2006). Self-Regulated Learning in
Technology-Enhanced Learning Environments: lessons
of a European peer review. European Journal of
Education 41, 353–379.
Woolf, B.P. (2010). Building intelligent interactive tutors:
student-centered strategies for revolutionizing e-
learning. Morgan Kaufmann.
CSEDU 2022 - 14th International Conference on Computer Supported Education
188