Embodied conversational agents as an interface for cognitively impaired older adults

Get Complete Project Material File(s) Now! »

Disabilities in older adults, loss of autonomy and exclusion

Thanks to the progress of health care in the past century, life expectancy at birth has dramatically increased in developed countries. But this comes at a cost: according to predictions of the United Nations, the proportion of people over 60 years old will represent over 20% of the population worldwide and over 35% in developed countries by 2050. The economic cost of aging demographics is huge: in the European Union, public spendings towards elderly people represented 25% of the GDP or about 50% of government expenditures in 2015, and it is expected to grow by more than 4% of the GDP, every year, until 2060 [62].
Though many people experience a healthy “normal” aging phenomenon, others are less lucky and develop severe age-related pathologies, resulting in disability. According to [159] (pp. 14-15), three models of disability are frequently used in the related literature: a medical model, a social model and a bio-psychosocial model. From a medical point of view, a disability can be defined as “a direct consequence of a disease, trauma or other health condition, which requires medical care in the form of individual treatment provided by health professionals” [159]. The social model “considers disability as a socially created problem and not at all an attribute of an individual” [159]. Lastly, the bio-psychosocial model was proposed in the International Classification of Functioning, Disability and Health (ICF) [225]. It takes both biological and social factors into account. Within this model, the term disability may refer to either an impairment in a body function or structure, a limitation to perform daily activities or a restriction to participate in everyday life situations. Given this latest definition of disability, the number of people in this situation grows rapidly with age, from 6.4% of the population in the 18-50 years old age group to 29.5% in the 60 and older age group, in high-income countries [226]. Another way to look at disability is to consider what kinds of daily activities the person’s conditions prevent him or her to perform. Daily activities may be divided in two groups: instrumental activities of daily living (IADL) and basic activities of daily living (BADL). IADL include management of personal finances, meal preparation or use of public and private transportation, whereas BADL include eating, drinking, dressing or bathing. Of course, people who are not capable of performing either of these groups of activities are not able to live independently, but people who cannot perform BADL require more care and support than people who are only prevented from performing IADL [123].

Truly adapted technologies to achieve good usability and acceptance

One of the main issues faced by assistive technologies is their adoption by older adults. Indeed, many older adults that could benefit from assistive technologies do not use them. There are a number of reasons why: lack of available information about the existing technologies that could help; some devices are too expensive for many people; using assistive technologies is considered as stigmatizing; people do not want to use a device because they do not understand it and are afraid they will not be able to use it; etc. As a result, many older adults do not get equipped with devices that could improve their quality of life. Once people know about a device that could help them and that they can afford, its adoption depends on two factors: usability and acceptance.
For a technology to be usable it has to be easy enough for the target users to use it. This condition may seem easy to fill but usability design issues can prove quite tricky as soon as people get old and/or have some form of impairment, as it will be explained in the next chapter. Typically, personal computers are easy to use for people who were born after their invention and with no disability, but visually impaired people, even young ones, cannot use them unless special arrangements are made. Older adults may struggle learning how to use computers, because they are so complicated, and, though touchscreen computers are a more user-friendly alternative that many older adults have adopted, people with cognitive impairment will still struggle learning the procedures to connect the device to the Wi-Fi, download applications, use each application that they have downloaded, etc. This is why devices and software have to be tailored to the special needs of these people, the ideal case being a device that requires no learning and with a shape that clearly suggests what it is for. In that regard, Embodied Conversational Agents (ECA), such as the one studied in this thesis, seem like a good way of interacting with computers for older adults with cognitive impairment: they look like people (or humanoids) and talk, something people have done their whole life, which is natural and easy to understand. In addition, they can be active and start addressing people to engage them, which is an advantage, as people with dementia often have a deficit for initiating actions.

Challenges in designing for older adults with special needs

It is a common mistake to consider memory loss and disorientation as part of the “normal” aging process when, in fact, these symptoms of cognitive impairment are caused by brain diseases, such as Alzheimer’s disease. The purpose of this section is to identify the physical, sensory and cognitive limitations experienced by older adults in normal aging and in the pathological cases addressed in this work, namely cognitive impairment and post-fall syndrome (PFS).
As already stated in the previous chapter, it is worth keeping in mind that older adults are often affected by more than one pathology. In fact, part of the work of physicians in geriatric care is to decide which pathologies should be treated in priority and to limit as much as possible undesirable side-effects caused by the interactions between all pharmacological treatments received by patients. In our case, PFS and cognitive impairment are not mutually exclusive, and can even be accompanied with other pathologies, such as hearing impairment or osteoarthrosis.


Table of contents :

Handicap chez les personnes âgées
Troubles cognitifs et démence chez les personnes âgées
Chutes chez les personnes âgées
Promesses des technologies d’assistance
Technologies pour le soin et l’assistance
Des technologies réellement adaptées pour l’utilisabilité et l’acceptabilité
Des technologies pour compléter l’aide humaine et non la remplacer
Agents conversationnels animés comme interface Homme-machine
pour les personnes âgées ayant des troubles cognitifs .
Thérapie par exposition en réalité virtuelle pour le traitement
du syndrome post-chute
Structure de la thèse
1 Introduction 
1.1 Disabilities in older adults
1.1.1 Cognitive impairment and dementia in older adults
1.1.2 Falls in older adults
1.2 Promises of assistive technologies
1.2.1 Technologies for care and assistance
1.2.2 Truly adapted technologies to achieve good usability and acceptance
1.3 Goals
1.3.1 Technology that complements human help but does not replace it
1.3.2 Embodied conversational agents as user interface for older adults with cognitive impairment
1.3.3 Virtual reality exposure therapy for post-fall syndrome
1.4 Contributions
1.5 Organization of this thesis
2 Game technology for older adults 
2.1 Design challenges
2.1.1 Normal aging
2.1.2 Cognitive impairment and dementia
2.1.3 Post-fall syndrome
2.1.4 Medical applications and hospital environment
2.1.5 Discussion
2.2 Game technology for elders’ care
2.2.1 Embodied conversational agents as an interface for cognitively impaired older adults
2.2.2 Other Applications of ECAs in elderly care
2.2.3 Games for fall prevention and rehabilitation in older adults
3 Focus points and methodology 
3.1 Key Concepts Involved in this Approach
3.1.1 Affordance
3.1.2 Attention
3.1.3 Engagement
3.1.4 Motivation
3.2 Living lab participatory design
3.2.1 Principles
3.2.2 The living lab: a place and a methodology
3.2.3 The Broca Hospital’s living lab
3.2.4 Off-the-shelf hardware, software and graphical elements
4 Creating virtual humans 
4.1 Human communication and SSP
4.2 Embodied conversational agents design
4.3 User behavior analysis
4.3.1 Automatic speech recognition
4.3.2 Attention and Engagement
4.3.3 Emotion recognition
4.4 Artificial social behavior generation
4.4.1 Behavior realizers
4.4.2 Speech synthesis
4.4.3 Evolving human-agent relationships and personality models
4.5 Dialog management
4.6 Domain-specific languages in ECA
4.6.1 XML 1Thanks to Anne-Sophie Rigaud for her help in writing this section.
4.6.2 BML
4.6.3 FML
4.6.4 PML
4.6.5 SSML
4.7 Discussion
5.1 Goals and design approach
5.2 Phase 1: Wizard of Oz study
5.2.1 Prototype description
5.2.2 Attention estimator
5.2.3 Experiment protocol
5.2.4 Performance of the attention estimator
5.2.5 Qualitative results
5.2.6 Anthropological analysis of the interactions2
5.2.7 Discussion and impact on the next design iteration
5.3 Questionnaires and focus groups
5.3.1 Questionnaires
5.3.2 Focus group
5.3.3 Discussion
5.4 Phase 2: Automating LOUISE
5.4.1 System overview
5.4.2 Behavior analysis
5.4.3 Interaction manager
5.4.4 Behavior realizer
5.4.5 Scenario description language
6 Assisted task management using LOUISE 
6.1 Goals
6.2 Structures of conversation
6.2.1 Dialog initiative
6.2.2 Asking questions and reacting to answers
6.2.3 Multiple choices
6.2.4 Step-by-step task instructions
6.3 LOUISE use scenarios
6.3.1 Drinking water
6.3.2 Taking pills
6.3.3 Measuring blood pressure
6.3.4 Choosing the menu of a meal
6.4 Experiment Protocol
6.4.1 Participants
6.4.2 Data analysis
2Thanks to Giovanni Carletti and Yann Laurent for their help in writing this section.
6.5 Results
6.5.1 Quantitative results and observations
6.5.2 Feedback
6.5.3 General questions about ECAs
6.6 Discussion
7 Virtual Promenade 
7.1 Goals and design approach
7.2 System description
7.2.1 VP game
7.2.2 Backwell moving seat
7.2.3 Game controllers
7.3 Playtesting
7.3.1 Virtual environments, avatars and view
7.3.2 Game controllers
7.3.3 Haptic chair and immersion
7.3.4 Tutorial
7.3.5 Discussion of design choices
7.3.6 Validation of the first version
7.4 Focus groups and co-design with care professionals
7.4.1 Feedback from the physiotherapists
7.4.2 Feedback from the psychomotricians
7.4.3 Discussion and improvements
7.5 Shadowing physiotherapists
7.6 Discussion
7.7 Pilot study in ecological settings
7.7.1 Protocol
7.7.2 Results
7.7.3 Discussion
8 Conclusion 
8.1 Case studies for virtual humans
8.1.1 LOUISE
8.1.2 Virtual Promenade
8.2 Living lab-based design for older adults
8.3 Future work
8.3.1 Conducting more validations studies
8.3.2 LOUISE
8.3.3 Virtual Promenade
8.3.4 Combining both approaches of virtual humans
8.4 An ethical epilogue


Related Posts