Designing Accessible User Interfaces for All by Means of Adaptive Systems

Accessible user interfaces are useful for people from a population with the widest range of capabilities. People with severe impairments primarily benefit from assistive technology while built-in software-based accessibility functionality and its customisation is advantageous to many other people who experience temporary or situational disabilities. However, increasing software customisation does not naturally result in better user interfaces or reduces barriers. Quite the reverse! Finding proper adjustments requires high computer literacy. Moreover, users must create mental models for different user interfaces and must be able to translate between them back and forth because several digital devices are used sequentially or simultaneous for multifaceted contexts. This thesis investigates accessible design for multifaceted usage of Information and Communication Technology (ICT) such as desktop, mobile devices, web-based applications or ubiquitous services to address people from a population with the widest range of capabilities. Particular attention is given to user-adaptive systems because they can overcome the limitations of manual forms of personalisation and thus can lead to better user interfaces for all people. The presented work is organised in three parts. The first part elaborates the foundations of accessible design. A distinctive domain of a user-adaptive systems called Adaptive Inclusive Interactive System (AIIS) is introduced to show personalisation embodies a new form of accessible design. AIISs distinguish from other user-adaptive systems because they take disability as a starting point for user modelling. Against the background of existing classifications, the common ground between AIIS and other domains of user-adaptive systems is elaborated and distinctive features of AIIS are discussed. The adaptor type of inference is newly introduced in order to describe use-adaptive systems adapting a single application and those considering the interplay of accessibility aids available on an interactive device. More specifically, the interaction within and across different configuration layers (i.e.\,operating system, application, assistive technology, web) is taken into account by inferring from existing knowledge to customisation for new platforms. Furthermore, functional requirements and design choices for AIISs are demonstrated along with five specific use cases; in particular respecting information appliances in ubiquitous or individual environments, highly customisable general purpose computers, multi-user application contexts and multi-screen behaviour. In addition, a systematic literature review of AIISs was conducted to discuss limitations of current approaches and to analyse the degree of user diversity targeted by existing systems. Results show, although multi-device and context-sensitive approaches are widely addressed, identified challenges are not sufficiently treated. Results of the literature survey show further, the adaptation theory of AIISs is mostly based on stereotypical assumptions about commonly known groups of impairments. A main limitation of stereotypes is that within-group diversity cannot be modelled. Feature-based user models such as needs and preferences are promising but seldom applied by AIISs. The second part of this thesis describes a knowledge-based approach to user interface adaptation from preferences and for special needs. The approach was implemented as the Rule-based Matchmaker (RBMM) component of the Global Public Inclusive Infrastructure (GPII). The GPII is a software architecture to facilitate the auto-configuration of ICT such as desktop computers running MS Windows or Linux, Android-based mobile devices or public terminals such as automated teller machines. Compared to the state of the art, the inference of the proposed knowledge-approach is not encoded into explicit rules deriving adaptations from impairment groups by propositional logic. Instead, a domain ontology and logical assertions were formalised to allow inferences compatible with those applied by accessibility experts who set-up interactive devices for people with disabilities. This knowledge was described in a generic and feature-based manner to ensure scalability of the inference about diverse user demands and heterogeneous accessibility aids. The formalisation of the domain knowledge and the logical rules fulfil several challenges, including deducing configurations in cases where a user need cannot be satisfied by device customisation or can be satisfied by a multitude of accessibility aids that interfere with one another. As the proposed system emulates decision-making of accessibility experts, automatically deduced configurations were validated against manual configurations of ten accessibility experts. Results show, the average matching score of the developed system is high. Thus, the proposed system can be considered being capable of making precise decisions towards personalising user interfaces based on user needs and preferences. Third part of this thesis comprises three empirical studies in order to discover specific domain knowledge with respect to insufficiently investigated AIISs-specific use cases as well as ICT-related requirements of certain user groups. Findings of this part of the thesis can be put to partial use -- for instance -- by extending the domain ontology of the proposed knowledge-based approach to user interface adaptation from preferences and for special needs. The first foundational user study investigates what people with disabilities know about their needs and preferences. The results show, user models must include ICT-related difficulties described at a more abstract level in addition to needs and preferences. Findings also demonstrate a wide range of individual requirements that must be considered by AIIS. Knowledge obtained from this user study was primarily used to derive requirements for the implemented RBMM. The second user study addresses a specific use case of AIIS, one that targets multi-user applications. More specifically, contradicting preferences in multi-user application contexts were analysed. Results comprise conflict resolution approaches that can be applied and further investigated by user-adaptive systems targeting this specific use case. A third user study explores ICT-related requirements of people with dementia because cognitive disabilities are not adequately covered by accessible design. The findings indicate an intersection of requirements with existing accessibility guidelines but they emphasise aspects not covered previously