=Paper=
{{Paper
|id=None
|storemode=property
|title=TERENCE: An Adaptive Learning System for Reasoning about Stories with Poor Comprehenders and their Educators
|pdfUrl=https://ceur-ws.org/Vol-810/paper-s05.pdf
|volume=Vol-810
|dblpUrl=https://dblp.org/rec/conf/cilc/MascioGV11
}}
==TERENCE: An Adaptive Learning System for Reasoning about Stories with Poor Comprehenders and their Educators==
https://ceur-ws.org/Vol-810/paper-s05.pdf
TERENCE: An Adaptive Learning System for
Reasoning about Stories with Poor Comprehenders
and their Educators
Tania di Mascio1 , Rosella Gennari2 , and Pierpaolo Vittorini3
1
DIEI – University of L’Aquila – Via G. Gronchi – 67100, L’Aquila, IT.
2
CS Faculty, Free University of Bozen-Bolzano - P.za Domenicani, 3 – 39100 Bolzano, IT.
3
MISP – University of L’Aquila – P.le S. Tommasi, 1 – 67100 Coppito, L’Aquila, IT.
Abstract. Text comprehension skills and strategies develop enormously from
the age of 7-8 until the age of 11, when children advance as independent readers.
Nowadays, more and more young children turn out to be poor (text) compre-
henders: they demonstrate text comprehension difficulties, related to inference-
making skills, despite proficiency in word decoding and other low-level cogni-
tive skills. Though there are several pencil-and-paper reading interventions for
improving inference-making skills on text, and specifically addressed to poor
comprehenders, the design and development of adaptive learning systems for this
purpose are lagging behind. The use of more intelligent adaptive learning systems
to tailor such interventions to poor comprehenders has a tremendous potential.
TERENCE embodies that potential. It is a Collaborative Project funded by the
EC under the ICT Call 5 FP7-ICT-2009-5 (1 October 2010-30 September 2013).
TERENCE will design and develop an intelligent adaptive learning system. In
particular, the system’s smart games will ask children to draw inferences about
events of stories, in Italian and in English. Moreover, the system will allow teach-
ers to choose and custom-tailor the types of stories and games according to the
needs of their learners. The TERENCE consortium involves European experts in
diverse and complementary fields (art and design, computer science, engineering,
linguistics, evidence-based medicine, psychology), and sees the constant involve-
ment of the end-users (poor comprehenders and their educators) from schools in
the UK and in Italy. The paper overviews the TERENCE project.
1 Introduction
Developing the capabilities of children to comprehend written texts is key to their de-
velopment as young adults. From the age of 7-8 until the age of 11, children develop as
independent readers. Nowadays, more and more children in that age range turn out to
be poor (text) comprehenders: they demonstrate difficulties in deep text comprehension,
despite well developed low-level cognitive skills like vocabulary knowledge, e.g., see
[2] for hearing poor comprehenders, and [5] for deaf poor comprehenders. In particular,
several studies experimentally demonstrate that poor comprehenders fail to master the
following reasoning skills in processing written stories:
(s1) coherent use of cohesive devices such as temporal connectives,
(s2) inference-making from different or distant parts of a text, integrating them coher-
ently,
(s3) detection of inconsistencies in texts.
Nowadays, there is clear evidence that such reasoning skills (s1, s2, and s3) are very
likely to be causally implicated in the development of deep text comprehension. In
particular, experiments show that inference-making questions centred around (s1, s2,
and s3), together with adequate visual aids, are pedagogically effective in fostering
deep comprehension of stories, e.g., see [4].
However, finding stories and educational material that are appropriate for poor com-
prehenders is a challenge, and hence educators are left alone in their daily interaction
with poor comprehenders. Most learning material for novice comprehenders is paper
based, and is not easily customisable to the specific requirements of poor comprehen-
ders, e.g., in the types, number or position of temporal connectives. Few systems pro-
mote general reading interventions, but they have high-school or university textbooks
as learning material, instead of stories, and are developed for old children or adults, and
not specifically for poor comprehenders.
TERENCE is a Collaborative Project funded by the EC under the ICT Call 5 FP7-
ICT-2009-5 (1 October 2010-30 September 2013) and aims at designing and developing
an intelligent adaptive learning system (ALS) [1]. The main end-users of the TER-
ENCE ALS are: learners, namely, primary-school poor comprehenders, hearing and
deaf, older than 7-8; educators, namely, primary-school teachers and support teachers,
as well as parents of the TERENCE learners. The learning material (in English and in
Italian) of the TERENCE ALS will be stories adapted to the specific requirements of
poor comprehenders, and its reading interventions will be mainly interactive question-
games centred around reasoning skills, like (s1), (s2), and (s3) above, that foster the
development of deep text comprehension, both accompanied by adequate visual aids.
This paper outlines the TERENCE project and the work already conducted therein.
It first specifies the chosen design methodology in Section 2. Then it outlines the con-
ceptual model of the TERENCE ALS in Section 2. It continues with the design of the
architecture in Section 4 and ends with a recap conclusive section.
2 The Design Methodology
The TERENCE system is developed following the user-centred design (UCD) method-
ology [3]. Generally speaking, the UCD places the end user, actively, at the centre of
the design process, which is iteratively repeated until attaining the usability of the sys-
tem. The iterative process revolves around the following main activities: a) analysis and
specification of the context of use; b) analysis and specification of the user require-
ments; c) design prototypes; d) evaluate the prototypes against the specification of the
requirements. In TERENCE we went through the first phase of the analysis of the con-
text of use and the requirements, which resulted into the semi-formal specification of:
1. the characteristics of the users, like knowledge, skills, experience, education, train-
ing, physical attributes, habits and capabilities;
2. the tasks, like successful reading interventions by class teachers for improving read-
ing comprehension;
3. the environments, divided into organisational, physical and socio-cultural charac-
teristics that may influence the usage and acceptance of the system.
Our design of the conceptual model of the TERENCE ALS, which is outlined in Sec-
tion 3 below, is based on such specifications and on the study of the state of the art
of conceptual modelling for ALSs. In the UCD, the design evolves cyclically through
refinements of the requirements, thus also the conceptual models of TERENCE will be
updated accordingly. The following section outlines the current conceptual model.
3 The TERENCE Conceptual Model
As mentioned above, TERENCE is developed as an ALS. Its conceptual model in-
cludes:
1. the domain model for the learning material, that are stories and games,
2. the user model for the users, including the educator and learner sub-models,
3. the adaptation model for the adaptation learning process.
In the remainder, for space limitations, we only sketch the design of the domain model,
and of the learner sub-model of the user model.
3.1 The Learner Sub-Model
The study of the state of the art for user modeling in ALSs and the specifications of
the user characteristics, in particular, the poor comprehenders’ reading skills, allowed
the consortium to define the following sub-models of the user model: the learner sub-
model; the educator sub-model; the expert sub-model.
The learner sub-model consists of three parts. The first structures general basic data
about the learner such as name. The second is really specific to TERENCE: it structures
information concerning the reading comprehension skills of poor comprehenders, anal-
ysed and specified in semi-formal format by the TERENCE consortium. By analysing
such skills, the system will be able to adapt, to its learner, its learning material, struc-
tured as in the domain model. The third part of the learner sub-model structures data
about the learner’s interaction with the system.
3.2 The Domain Model
The main learning material of the TERENCE ALS consists in illustrated stories and
games. During the analysis of the requirements and of the context of use of TERENCE,
the tasks of the users were analyses and specified. Their specification allowed us to
structure the learning material as in the domain model. This is divided into the story
and game sub-models.
The story model represents data related to the stories that the learners interact with.
For instance, stories are structured into books, each book has a genre and related avatars.
Furthermore, each story is annotated using natural language processing tools. The an-
notations, for instance, provide us with the events of which the story is made, specifying
their actors, location and temporal relations.
Each story has associated games. Preliminary prototypes of the games are being
designed by following assessed interventions for stimulating the reasoning skills on
texts outlined in the introductory section and resulting from our user requirements’
semi-formal analysis. For instance, according to this, poor comprehenders are in need of
inference-making games that pose and solicit questions about relations between events
in the stories, monitoring the learners’ comprehension of the story flow, e.g., “Does the
big eggshell crack before Mummy Duck watches it?”. The game model represents the
data for the TERENCE games, which are divided into smart games and relaxing games.
Smart games are centered around inference-making questions that are built upon the
annotations of the stories. As such, smart games are taxing for poor comprehenders and
need to be paused by relaxing games (e.g., draw your favourite character), which keep
the learners’ attention and motivation alive. More details concerning their generation
and resolution are in the following section.
4 The Architecture of the TERENCE System
The current logical architecture of our ALS is divided into three layers, namely, the data
layer, the application layer, and the interaction layer. The data layer stores, for instance,
the stories and games according to the domain model, and the information specific to
each learner according to the learner model.
The application layer implements the adaptation engine, and the intelligent back-
bone responsible for the feedback on games (e.g., correct or incorrect answers to games).
Finally, the interaction layer contains the users’ interfaces.
The main modules of the adaptation engine of our ALS are a constraint-based auto-
mated reasoner and a natural language processing (NLP) module, consisting of a pro-
cessor for English stories and one for Italian stories, that lay at the core of the adaptation
engine and constitute its intelligent backbone, so to speak. The NLP modules serve pri-
marily to annotate stories with specific XML tags. The tags are used for classifying
stories as in the story sub-model, as well as for generating and giving feedback on the
TERENCE smart games. According to the assessment of the students’ performance on
a class of games, the adaption engine will attempt to guide the students to the most
adequate games and stories, following the adaptation model.
In the specific context of our ALS, tasks will be implemented through web services
and their composition. For instance, let us refer to the generation of a temporal question-
game centred around the question “Does the big eggshell crack before Mummy Duck
watches it?”. The composition will go as follows:
1. firstly, the system invokes an NLP web service operation, which takes the story as
input and returns the annotations in the novel annotation language;
2. secondly, these annotations are taken as input, and an operation of the Automated
Reasoner service deduces further temporal annotations as output, updating them;
3. finally, the updated annotations are taken as input and a further operation of the
NLP web service generates as output the grammatically correct question-game.
A first clear advantage of such an approach is that it allows for the reuse of our architec-
ture in other languages, by implementing the appropriate NLP services. Furthermore,
since web services are accessible through HTTP calls, they can be invoked directly in
their respective organisations, e.g. for keeping protected any eventual patent. Finally,
the high-level operations might also be implemented with a programming-in-the-large
paradigm, e.g. BPEL, thus allowing for an easy deploy of further operations, that be-
come web services, and therefore re-usable to build up more complex tasks.
5 Conclusions
TERENCE is European project for the design and evaluation of an ALS specific for
poor comprehenders and their educators. More in general, the project aims at offering
innovative usability and evaluation methodologies, pedagogical models, AI technolo-
gies, and an ALS for reasoning about stories through smart games, in Italian and in
English, all developed via a coordinated and cross-disciplinary effort of European ex-
perts in diverse and complementary fields (art and design, computer science, engineer-
ing, linguistics, psychology), and with the constant involvement of the end-users (poor
comprehenders, deaf children and their educators) from schools in the UK and in Italy.
Acknowledgments. The authors’ work was supported by TERENCE project. TER-
ENCE is funded by the European Commission through the Seventh Framework Pro-
gramme for RTD, Strategic Objective ICT-2009.4.2, ICT, Technology-enhanced learn-
ing. The contents of the paper reflects only the authors’ view and the European Com-
mission is not liable for it.
References
1. P. Brusilovsky. Adaptive Hypermedia. User Modeling and User-Adapted Interaction, 11:87–
110, March 2001.
2. K. Cain and J. Oakhill, editors. Children’s Comprehension Problems in Oral and Written
Language: A Cognitive Perspective. Guildford Press, 2007.
3. J. Gulliksen, B. Göransson, I. Boivie, S. Blomkvist, J. Persson, and A. Cajanger. Key Princi-
ples for User-centred Systems Design. 22(6):397–409, 2003.
4. National Reading Panel. Teaching Children to Read: An evidence-based Assessment of the
Scientific Research Literature on Reading and its Implications for Reading Instruction. Tech-
nical report, National Institute of Child Health and Human Development, 2000.
5. P. Spencer and M. Marschark. Evidence-based Practice in educating Deaf and Hard-of-
hearing Students. Oxford University Press, New York, NY, USA, 2010.