Personal Learning Environment for Disabled People Diana Janeth Lancheros-Cuesta Universidad de La Salle Pontificia Universidad Javeriana Departamento Ingeniería de Sistemas Bogotá, Colombia
[email protected]
Angela Carrillo-Ramos,
Jaime Pavlich-Mariscal
Pontificia Universidad Javeriana Departamento Ingeniería de Sistemas Bogotá, Colombia
[email protected]
Pontificia Universidad Javeriana Departamento Ingeniería de Sistemas Bogotá, Colombia
[email protected] Universidad Católica del Norte Departamento de Ingeniería de Sistemas Antofagasta, Chile
Abstract—Learning difficulties may affect the development of cognitive abilities, text interpretation, reading-writing, among others. Therefore, different methods and evaluation techniques are required to beincluded in virtual learning environments. This paper presents a Personal Learning Environment for People with Disabilities, a system that adapts information display in an elearning environment, according to characteristics and difficulties of a student with mild cognitive disabilities. The Personal Learning Environment for People with Disabilities includes a neural network to classify disability type (sensory orcognitive), which becomes a base to build virtual environments that can adapt and personalize information to students, taking into account their context and profile.
adapts the information of e-learning services to students. Figure 1 describes the main components of this environment.
Keywords-disabilities, learning environments, neural networks, adaptation, personalization
I.
INTRODUCTION
Colombian Law 1145 defines a disabled person as somebody who “has limitations and deficiencies in daily activities and constraints to social participation, due to a health condition or physical, environmental, cultural, social, or day-to-day barriers.”[8]. Among the multiple ways to assist people with disabilities, software systems may be of great help, since they can dynamically convey the required information and process and analyze the response of their users to improve interaction. Many systems developed with this purpose have focused on sensory and motor disabilities [13], while fewer works have focused on cognitive disabilities [18]. Therefore, it is necessary to develop more tools to identify, interpret, and address the needs of people with cognitive disabilities. Among these tools for cognitive development there are three categories of interest: virtual learning environments (VLE), computer assisted instruction (CAI), and intelligent tutoring systems (ITS). There are two important improvement opportunities for these systems. First, artificial intelligence techniques can be used to predict people’s behavior and model their interactions in e-learning environments [25]. Second, the system could use the above information to adapt to the particular characteristics of each student, to provide him/her with the information that is most adequate for his/her learning requirements. The final goal of this research is to create a Personal Learning Environment for People with Disabilities, which
Figure 1. Personal Learning Environment Components.
The Student Profile (a) stores the student characteristics associated to learning style and preferences with respect to interaction with the system. The Disability profile (b) includes social and cognitive characteristics and also the main suggestions about information display for people with disabilities, either cognitive or physical. The Disability profile also has a Disability Classification Model that dynamically updates the Student Profile and classifies his/her disability, according to the interaction with the system. The Personal Learning Environment for People with Disabilities includes an Application Model that provides the data that the Personal Learning Environment for People with Disabilities uses to enrich two services: information retrieval and execution of activities in the learning environment. Based on these two services and the particular characteristics and information requirements of the student, the system selects the most adequate (VLO) Virtual Learning Objects. An expected result of Personal Learning Environment for People with
Disabilities is the creation of personalized courses. Those
courses would present specific topics to each student and would also be able to automatically generate learning activities specifically tailored for them. This paper is the first step towards the creation of a fullfeatured Personal Learning Environment for People with Disabilities and focuses on describing the Disability Model and its Disability Prediction Model, based on neural networks. The paper is structured as follows: Section II describes the main concepts associated to disabilities, adaptation, virtual learning environments, and neural networks. Section III presents related works. Section IV describes the Disability Model. Section V details the neural network technique utilized to predict disabilities. Section VI describes the application of the Personal Learning Environment for People with Disabilities in a case study oriented to Asperger students. Section VII concludes the paper and presents future work. II.
BACKGROUND
This section details the main concepts required to understand the proposed approach. Disability is “the quality of a disabled person”. The latter is defined as “a person who has impeded or obstructed one of his/her day-today normal activities, because of an alteration in his/her intellectual of physical functions”. The World Health Organization (WHO) classifies sensory and cognitive disabilities in four categories [13]: Sensory and communication disabilities, communication and language comprehension disabilities, motor disabilities, and cognitive disabilities. People with cognitive disabilities tend to have difficulties in performing day-to-day activities, such as personal hygiene, home activities, school duties, etc. This category includes disabilities such as Alzheimer, Down, and Asperger Syndromes, among others. Metacognition comprises the knowledge and regulation of cognitive activities associated to learning processes [12]. Metacognition separates metacognitive knowledge from abilities. The former is the declarative knowledge about the person, his/her interactions with others, tasks, and strategy. The latter is associated to procedural knowledge to regulate problem-solving and cognitive activities [17]. A pedagogical support model for collaboration in computational environments [10] is utilized to encourage the student to reason and contribute to the learning environment, also developing metacognitive abilities. Messages written by students in forums encourage reasoning and reflection, and become characteristics of metacognitive processes [12]. The student can decide to participate in the debate, asking questions, asking for help, or providing answers [1]. In contrast to passive participation, active participation in online learning requires a different type of tracking and control. For instance, explanation construction [22] tries to motivate students to become aware of their thinking process. During online discussion, it is also essential that students present their ideas as clear as possible, so they can be understood by the other participants, and students could develop their metacognitive knowledge [12]. Kardan et al. [9] define adaptation as “the action of accommodating, adjusting something to another thing”. They indicate that adaptative systems are different than adaptable systems. The main difference lies in the part that is responsible of the adaptation process. If the user is responsible for the adaptation process,
then the system is adaptable; if the system adapts to the users’ necessities, then the system is adaptative. To perform the adaptation process, the adaptive system must automatically collect specific information about its users and the environment. Min et al. [16] indicate that adaptative learning uses two main technologies: adaptative hypermedia systems, and semantic web. The former adapts navigation and content presentation. The latter adapts information using knowledge representation and reasoning rules processing. In the Personal Learning Environment for People with Disabilities (see Figure 1) the Interface Model, the Pedagogical Model, and the Domain Model can be automatically updated using artificial intelligence (AI) techniques. These techniques also allow to represent and reason about students’ knowledge [25]. The most used AI techniques are: formal logic, expert systems, planning methods, and artificial neural networks [23]. Artificial neural networks (ANN) often referred simply as “neural networks” is the technique being used in this paper. An ANN is based on the structure of the nervous system, mainly the brain. The main characteristics of ANN are: (a) The ability to learn from training data that includes the inputs and expected outputs; (b) The internal representation of information that does not require a knowledge base; (c) Redundancy in information storage,which allows an ANN to function even after it is partially damaged; (d) Flexibility in treating the input information, the response does not change significantly in presence of noise or other elements. III.
RELATED WORKS
There are two categories of related works: information systems to intervene and analyze Asperger Syndrome and Artificial Intelligence techniques applied to classify cognitive disabilities. Autism intervention includes technologies such as: virtual reality [3], simulation-based systems [5], and virtual learning agents [18]. In several studies, researchers use virtual animated characters [20]. Others have created virtual reality systems to familiarize people with Autism Spectrum Disorder (ASD) through social environments [24]. The REACT system assists the creation of interactive and personalized social scenarios for teenagers with mild autism [4]. Kuhlen [11] developed the Iris Graphics library, which uses virtual reality techniques to help peoples with disabilities. Kuhlen also describes a way medics can use virtual reality as an advanced tool to diagnose physical disabilities. Liu [14] presents a strategy to improve performance of students. The approach is to use statistical analysis and data mining techniques to predict learning styles and develop personalized strategies in the classroom. Menzies [15] developed a learning environment to improve knowledge and social interactions in Asperger students. Ozdemir [21] developed an information system that adapts information to autism children in different contexts, such as home or school. Morris [19] designed an algorithm to retrieve images in a system for Asperger students. The system facilitates personalization through the integration of images of interest into a device and the child activities. Hirano [7] demonstrated that visual schedules, i.e., the use of symbols to represent
different activities, have demonstrated to be an effective intervention technique to help people with Autism Spectrum Disorder. Anumolo [2] developed a system that uses a multilayered neural network to implement memory strategies. Those strategies are use to evaluate memories of children with mild mental retardation. The system simulates the strategies used by children to remember things and uses them to identify objects and patterns. This model simulates the object identification process of children. The neural network learns successful case patterns, which are later used in other children. One of its main advantages is the variability in strategy selection. For instance, in one simulation, researchers observed that the neural network selected a different strategy than the expected one. On the other hand, the neural network has its limitations. As children grow, their object identification strategies change. This neural network only implemented three strategies and it does not have mechanisms to adopt new strategies.
the way students should receive the information from the system. The types of suggestions are: (a) Format of the information display, according to the identified disability (e.g., sounds and voices for visually-impaired students, images and graphics for hearing-impaired students); (b) access device, such as mouse, Braille keyboard, speech recognition software, among others; (c) perception of information, to determine the best way a student comprehends the information provided by the system; and, (d) Navigation, which indicates the best way to navigate the information, according to the disability. In addition, the Disability Model includes learning aspects: memory, attention, and language. Memory has two classifications: (a) Short-term memory, associated to information that needs to be recorded for a few minutes or seconds; and, (b) Long-term memory, associated to information that needs to be recorded over long periods [11].
Ghosh-Dastidar [10] developed a neural network to precisely diagnose Alzheimer’s disease. The neural network has three hidden layers and uses supervised learning. The approach is for individuals with the disease and healthy individuals, based on the information of the central auditory nervous system. The main advantage of this approach is that it is very precise to diagnose Alzheimer’s disease and the diagnostic is consistent with the advancement of the disease. IV.
DISABILITY MODEL FOR LEARNINGENVIRONMENTS
Figure 2 shows the inputs and outputs of the process to generate the disability profile. There is a data repository with the disability profile, which includes characteristics, suggestions about the information system and learning aspects that are affected in the teaching-learning process. This information is provided by the teacher, based on the information disability experts provide.
Teacher Disability
Details disability
Aspects of learning
Generate disability profile Disability Profile
Detail aspects of learning
Figure 3. Disability Model for Learning Environments. [26]
Suggestions Detail of the suggestions
Figure 2. Generate disability profile.
This section details the proposed Disability Model (DM) for learning environments. Figure 3 is an overview of the model [26]. Given the preferences and the student behavior, the DM allows to suggest the most adequate virtual learning objects for that student. The DM includes information about student disabilities and performance evaluation aspects, such as memory, attention, and language. Each disability includes a name, social and cognitive characteristics, difficulties to access the system, and sensory organ affected by the disability. For each disability, the model provides different suggestions, i.e.,
Attention includes information about student focus on tools and activities of the learning environment. Language indicates whether the student comprehends, interprets, or utilizes language in the learning environment. Skills and performance of cognitive activities define a specific profile, depending on the disability. For instance, a person with hearing disabilities may present problems in language comprehension, thus affecting the communication aspects of his/her profile. The model allows to identify difficulties in the interaction with the system due to disabilities. For instance, for physical disabilities, interaction difficulties arise due to the user being unable to use some access devices. For cognitive disabilities, difficulties are reflected in the lack of alternative forms of information display. Attention problems may be reflected in visual or auditory distractions that cannot be disabled, lack of a clear organization of the material, among others. For sensory disabilities, such as hearing impairment, difficulties may arise
due to the lack of visual material. Similarly, for visual disabilities, difficulties may arise due to the lack of audio material. V.
GENERATE DISABILITY PROFILE
The process shown in Figure 2 takes as input the data from the disability repository, learning aspects and suggestions. This information is feed by the teacher, taking into account the information provided by the disability expert. Figure 4 shows the algorithm to generate the disability profile.
To classify disabilities, the approach is to utilize a multilayered feed-forward neural network. Figure 5 shows the structure of the neural network. The input layer has four nodes, one hidden layer with three nodes, and one output layer. The input data is a vector with four data, each with a value between 0 and 1, each data means the following: (X1) how easy is verbal or written communication for the student. 0 means that the student has high difficulties 1 means that is not difficult for him/her to communicate. (X2) The student prefers virtual learning objects containing sounds and simulations. 0 means no preference, 1 means high preference. (X3) The student has difficulties interpreting social situations. 0 means no difficulties, 1 means high difficulties. (X4) The student prefers images of his/her favorite characters. 0 means no preference, 1 means high preference. The output layer has one node that outputs discrete values 0 or 1. 0 means that there is cognitive disability, 1 means that there is sensory disability. The neural network was trained and validated using Mathlab. The training dataset has 50 elements, while the test dataset has 30 elements.
Figure 5. Neural Network Structure
VI.
CASE STUDY
The Asperger Syndrome is an autism spectrum disorder (ASD) characterized for the inability to recognize and comprehend thoughts, beliefs, emotions, and intentions of other people. Typically, Asperger people have difficulties to comprehend and use non-verbal signs, such as facial gestures and voice inflection. There is also difficult to imagine social situations outside routine ones and to participate and feign in games [6]. During the development of this case study, the Asperger student characteristics were identified as mainly visual. Student preferred images with strong colors and medium sized fonts.
Figure 4. Algorithm "Generate Disability Profile"
The algorithm includes four processes. The first process creates the basic disability information, including social and cognitive characteristics. The second process defines format, device, perception and navigational suggestions that are better suited for each disability. The third process registers information about learning aspects that are affected by disabilities, e.g., attention, language, or memory. The last process creates the disability profile based on suggestions, characteristics and learning aspects, which are quantified and better suited for each individual. For instance, the process can generate several profiles of deaf students, which have different learning aspects. A deaf student may have memory difficulties will have a different disability profile than a deaf student with attention difficulties.
They also preferred simulations to learn social behavior (e.g., waiting in a library queue to borrow books). Navigation preferences are also visual. Students tend to follow navigation paths that include icons. The main cognitive processes of people with ASD that affect the teaching-learning process [24] are: (a) Perception. Visual and auditive perception performance is normal in people with ASD. In many cases, visual perception is superior, even in low-level activities (e.g., jigsaw puzzles). In auditory perception, absolute tone capabilities have been detected, even for isolated sounds learning aspect. (b) Memory. Procedural memory and motor tasks with serial reaction time, response times can be high. Sequential memory when performing tasks in a computer can have low results learning aspect - Memory. (c) Thinking. It is evaluated the capability to correctly solve tasks that require inferential knowledge over contents Profile. (d) Non-verbal communication. Interpersonal regulation characteristics (communication). Gestures, signs,
among others. Language. (e) Verbal communication. Prosodic voice characteristics, errors and distortion in articulating sounds, inadequate adjustment to phonetic models of the society, among others. Language. (f) Social relationship processes Low social abilities Disability characteristics. In addition, the Personal Learning Environment for People with Disabilities defines filters that allow enrichment of services. For instance, scenario modification, information display about social situations that are more adequate to the student needs. An example of this kind of filter is as in Figure 6. The disability profile lets to determine that Asperger people have problems with perception and output of information, language, attention, and memory, which affect the abilities of interpretation, comprehension and analysis of social situations. Sea fd ϵ Display_Format Sea r ϵ Activity Sea d ϵ Disability Filter_Activity(preference, r, d){ If (r == “Content”) { If (preference== ”Simulation” and d==”cognitive”) { Display(fd, device) } else { If (preference= ”Image” and d=”cognitive”) { Display(fd, device)}} }}} Display(fd, device, r) If (fdϵDeviceChar) { displayInfo(r) }else sustitution_format(r) }
Figure 6. Disability Profile Filter
Figures 7 and 8 show the displayadaptation in the Personal Learning Environment for People with Disabilities. Figure 7 shows a social situation in a queue, where the situation and rules to follow are explained to the child. The information display process takes into account the student profile and the Disability Model to deliver to the student the information as images that simulate a particular social situation and facilitate the development of thinking and social skills. Based on the classification scheme, the neural network would determine a cognitive disability, specifically for the Asperger syndrome.
Figure 8. Display Adaptation in the Personal Learning Environmentfor people with auditory disabilities
VII. CONCLUSIONS AND FUTURE WORK The Personal Learning Environment for People with Disabilities allows characterizing the student and a disability, determining the difficulties that can occur in the learning process in a virtual environment, to adapt information display. The application of the Disability Model allows determining that the greatest difficulties lie in the cognitive development aspects, such as language, attention, and memory. Similarly, there are difficulties in interpretation, analysis and comprehension abilities. The use of a neural network to classify, based on disability patterns, allows to update the Disability Profile and determine the adaptation suggestion in information display. Future work is to apply the prototype in other disabilities and perfecting the model to improve students’ performance in a virtual learning environment. In addition, system performance will be evaluated, based on the criteria defined by Lancioni et al. [13], which may contribute to the prediction of variables to use in information display. The main criteria are: time to change state in the system, work time per unit, time to complete a certain task, number of errors, error correction vs. number of errors, time to remember the required knowledge to perform certain tasks. The algorithm to generate the disability profile shows the necessity to create different disability profiles, depending on learning aspects. Future work also includes the creation of a prioritization algorithm to select the most adequate information to deliver according to the specific characteristics of the student, based on his/her disability profile. Finally, the system must be validated in a real environment with students with and without disabilities. REFERENCES
Figure7. Display Adaptation in the Personal Learning Environment based on theDisability Model of a student with Asperger Syndrome
On the other hand, if the neural network determines a sensory disability, such as auditory one, the adaptation in the information display would focus on sign language and simulations (see Figure 8).
[1] Aleven, V., McLaren, B., Roll, I., Koedinger, K.: Toward tutoring help seeking. In: Lester, J.C., Vicari, R.M., Paraguaçu, F. (eds.) Intelligent Tutoring Systems, Lecture Notes in Computer Science, vol. 3220, pp. 227–239. Springer Berlin / Heidelberg (2004) [2] Anumolu, V., Bray, N.W., Reilly, K.D.: Neural network models of strategy development in children. Neural Networks 10(1), 7 (1997) [3] Aranda, G., Vizcaino, A., Cechich, A., Piattini, M.: A cognitive perspective for choosing groupware tools and elicitation techniques in virtual teams. In: Computational
Science and Its Applications. Lecture Notes in Computer Science, vol. 3480, pp. 201–231 (2005) [4] Boujarwah, F.A., Riedl, M.O., Abowd, G.D., Arriaga, R.I.: React intelligent authoring of social skills instructional modules for adolescents with high-functioning autism. SIGACCESS Access Comput 1(99), 13 (2011) [5] Davis, M., Dautenhahn, K., Nehaniv, C., Powell, S.: Towards an interactive system eliciting narrative comprehension in children with autism: A longitudinal study. In: Clarkson, J., Langdon, P., Robinson, P. (eds.) Designing Accessible Technology, pp. 101– 114. Springer London (2006) [6] Duffy, C., Healy, O.: Spontaneous communication in autism spectrum disorder: A review of topographies and interventions. Research in Autism Spectrum Disorders 5(3), 977–983 (Jul 2011) [7] Hirano, S.H., Yeganyan, M.T., Marcu, G., Nguyen, D.H., Boyd, L.A., Hayes, G.R.: vsked evaluation of a system to support classroom activities for children with autism. In: Proceedings of the 28th international conference on Human factors in computing systems. pp. 1633–1642. CHI2010, ACM (2010) [8] Jimenez Bunuales, M., González Diego, P., Martín Moreno, J.M.: La clasificación internacional del funcionamiento de la discapacidad y de la salud (CIF) 2001. Revista Española de Salud Pública 76(4), 271–279 (Aug 2002) [9] Kardan, A., Monkaresi, H.: Developing a novel framework for effective use of implicit feedback in adaptive e-Learning. In: Information Technology: New Generations, 2008. ITNG 2008. Fifth International Conference on. pp. 955 –960 (Apr 2008) [10] Koschmann, T.D., Hall, R., Miyake, N.: CSCL 2, Carrying Forward the Conversation. Routledge (Mar 2002) [11] Kuhlen, T., Dohle, C.: Virtual reality for physically disabled people. Computers in Biology and Medicine 25(2), 205–211 (Mar 1995) [12] Kuhn, D.: Metacognitive development. Current Directions in Psychological Science 9(5), 178–181 (Oct 2000) [13] Lancioni, G.E., O’Reilly, M.F., Singh, N.N., Sigafoos, J., Oliva, D., Antonucci, M., Tota, A., Basili, G.: Microswitchbased programs for persons with multiple disabilities: An overview of some recent developments 1. Perceptual and Motor Skills 106(2), 355–370 (Apr 2008) [14] Liu, C.: A Simulation-Based experience in learning structures of bayesian networks to represent how students learn composite concepts. International Journal of Artificial Intelligence in Education 18(3), 237–285 (2008) [15] Menzies, R.: Promoting sharing behaviours in children through the use of a customised novel computer system. SIGACCESS Access Comput 54, 30 (2011)
[16] Min, W., Wei, C., Lei, C.: Research of ontology-based adaptive learning system. In: Computational Intelligence and Design, 2008. ISCID ’08. International Symposium on. vol. 2, pp. 366 –370 (Oct 2008) [17] Minnaert, A., Janssen, P.J.: The additive effect of regulatory activities on top of intelligence in relation to academic performance in higher education. Learning and Instruction 9(1), 77–91 (Feb 1998) [18] Mohamad, Y., Velasco, C., Damm, S., Tebarth, H.: Cognitive training with animated pedagogical agents (TAPA) in children with learning disabilities. In: Miesenberger, K., Klaus, J., Zagler, W., Burger, D. (eds.) Computers Helping People with Special Needs, Lecture Notes in Computer Science, vol. 3118, pp. 629– 629. Springer Berlin / Heidelberg (2004) [19] Morris, R.R., Kirschbaum, C.R., Picard, R.W.: Broadening accessibility through special interests: a new approach for software customization. In: Proceedings of the 12th international ACM SIGACCESS conference on Computers and accessibility. p. 171. ASSETS 2010, ACM, New York, NY, USA (2010) [20] Ortiz, A., Oyarzun, D., del Puy Carretero, M.: ELEIN: ELearning with 3D interactive emotional agents. In: Chang, M., Kuo, R., Kinshuk, Chen, G., Hirose, M. (eds.) Learning by Playing. Game-based Education System Design and Development, Lecture Notes in Computer Science, vol. 5670, pp. 294–305. Springer Berlin / Heidelberg (2009) [21] Ozdemir, S.: Using multimedia social stories to increase appropriate social engagement in young children with autism. Online Submission 7 (Jul 2008) [22] Ploetzner, R., Dillenbourg, P., Praier, M., Traum, D.: Learning by explaining to oneself and to others. Tech. rep. (1999) [23] Russell, S.J., Norvig, P.: Artificial intelligence: a modern approach. Prentice-Hall, Upper Saddle River, NJ, 3rd ed. edn. (2010) [24] Tuedor, M.: Universal access through accessible computer educational programs to develop the reading skills of children with autistic spectrum disorders. Universal Access in the Information Society 5(3), 292–298 (2006) [25] Woolf, B.P.: Building Intelligent Interactive Tutors: Studentcentered strategies for revolutionizing e-learning. Morgan Kaufmann Publishers Inc., San Francisco, CA, USA (2007) [26]D. Lancheros C. and A. Carrillo-Ramos, "Modelo cdaptativo para la caracterización de Dificultades/Discapacidades en un ambiente virtual educativo," Dyna, vol. 79, no. 175, pp. 52-61, 2012.