PhD AI-Enabled Digital Accessibility (ADA)

For our first PhD cohort, we invite proposals for innovative and original projects that align with the themes and outlines presented below: 

THEME 1: Meaningful image/video-to-speech/text accessibility through AI 

Outline: Automating audio description for short-form video: user preferences and priorities

Digital media often feature short-form video, including news or social media clips, animations. However, such video clips often lack audio description (AD) for people with sight loss, especially in user-generated content. While AI technology may have the potential to address this gap, current AI-generated descriptions of visual content lack in sophistication, nuance, and narrative coherence. Short video clips with low narrative complexity offer a starting point for experimenting with automated AD. We invite proposals that explore how AI-generated descriptions of visual content can be shaped to make short-form video accessible for people with sight loss. The project should involve eliciting user needs, preferences, and priorities, as well as experimenting with prototypes of AI-generated description to identify areas of challenge and direct further development. 

THEME 2: Meaningful speech-to-text accessibility through AI 

Outline: Mapping speech-to-text accessibility needs across diverse audiences

Speech-to-text (STT) services, such as subtitles and transcription, are vital for accessibility. 

Despite its increasing popularity on TV, video platforms and social media, implementing STT in real-time and into another language remains a complex problem, which cannot be addressed by technology alone. Fundamental questions need to be explored before personalised and context-specific solutions acknowledging diverse user needs can be developed. We invite proposals focusing on conducting audience reception studies with different types of user groups and/or digital media to identify relevant quality parameters to integrate the ones currently used (mostly accuracy and latency) as well as users’ expectations on a service that is meant to go cater for people with different needs. This project should complement and pave the way for more experimental, comparative studies across different workflows/configurations/digital materials. 

THEME 3: Improving text comprehension and simplification through AI 

Outline: Investigation of prompting techniques for text simplification

Large language models have been shown to be capable of simplifying texts when prompted to do so. However, in most cases the prompts are designed in an ad-hoc manner and do not effectively convey the needs of end users. We invite applications for a PhD project focused on developing a meta-language for writing prompts which can be used to facilitate text simplification. Ideally, this meta-language should be adaptable to any category of users, but the PhD project should focus on and involve specific user groups. Techniques such as prompting chain of thoughts and generating multimodal content may also be explored within this project, as long as they contribute to the development of the framework. 

THEME 4: Tools for meaningful cross-language ADA, especially low resource languages 

Outline: Identifying user needs for AI-enabled language access to digital health information 

Many healthcare institutions have turned to AI-driven solutions to make digital health information accessible to diverse linguistic groups. However, the language support provided on healthcare-related websites and by digital health apps, often through machine translation, may lack verification, leading to potential misinformation and miscommunication, especially for languages with underdeveloped machine translation capacity. We invite PhD projects investigating accessibility biases for speakers of minority languages in healthcare digital media. We encourage research adopting a user-centred approach to analysing and evaluating the use of machine translation of healthcare websites and apps. The project should involve speakers of minority/low resource languages to understand their needs and preferences, as a stepping stone towards enhancing language access and equality without introducing further risks. 

THEME 5: Enhancing assistive technology through AI 

Outline: Exploring preferences regarding AI-based solutions to accessibility of textual information for people living with dementia 

It can be difficult for people living with dementia to access information presented in complex texts. Existing guidance emphasises the importance of a clear layout, highlighting key points and careful use of images, but acknowledges that no single solution works for everyone. AI holds promise for assisting people living with dementia in accessing information, for example, by automatically adapting written text formats or shifting from written text to speech. AI offers the prospect of flexible solutions that adapt to changing needs. It is not clear, however, how AI-based accessibility tools can be flexibly deployed, and to what extent users would wish AI to make decisions on their behalf. This project would work with potential users to explore their insight into how AI solutions should acquire and act on user feedback to improve information accessibility for people living with dementia. We invite projects using qualitative research methods to build a design brief for AI-based textual accessibility. 

THEME 6: Developing novel methods to personalise ADA 

Outline (a): Benchmarking AI-enabled accessibility

The proliferation of digital media content has increased the demand for different modalities of audiovisual translation to ensure accessibility for users with diverse needs. Digital media providers are turning to AI-driven solutions to manage this demand and meet legislative requirements. Although achieving human-level sophistication through automated approaches remains challenging, initial explorations of AI-generated accessible content indicate potential benefits for users, provided minimum thresholds for accuracy and intelligibility and other aspects are met. Amidst the emerging blend of human, machine-assisted and fully automated methods for creating accessible content, user-centric evaluation is essential, regardless of production method. We invite projects that develop a unified framework for evaluating and benchmarking one specific type of accessible content (e.g., audio described content, or content adapted for neurodiverse people), with a view to extending this framework to other modalities in the future. 

Outline (b): ADA-Learn: Few-shot meta-interpretive learning for personalised digital accessibility 

Most current machine learning approaches for modelling user requirements and behaviour need large amounts of training data and the learned models are only suitable for the ‘average’ users (one-size-fits-all assumption). However, personalised digital accessibility requires machine learning algorithms that can build models from a small number of interactions with the user without the need for prior training or transfer learning (which could include implicit biases). We invite projects that aim to develop the theory and implementations for efficient learning from small data using Meta-Interpretive Learning (MIL) and in particular Meta-Inverse Entailment (MIE). MIE is a novel explainable few-shot learning approach developed at Surrey, which has been already used for learning accurate descriptions of complex concepts from single examples. Successful projects will extend MIE for explainable few-shot meta-interpretive learning for personalised digital accessibility.

Supervision 

As a PhD student, you will be supervised by a unique world-class team of researchers from different disciplines who are experts within a wide range of theoretical and methodological skills. Depending on your specific project, your supervisory team will be multi-disciplinary and you will benefit from the knowledge and insights from multiple Schools and Departments within the University, including the Centre for Translation Studies (CTS), the Centre for Vision, Speech and Signal Processing (CVSSP) and Nature Inspired Computer and Engineering Research Group (NICE). 

In addition, the Network will benefit from contributions by Surrey's Institute for People-Centred AI (PAI), Digital World Research Centre (DWRC), Digital Societies and Brain and Behaviour research groups. 

Your designated team of supervisors and co-supervisors, tailored to your specific project, will help you define the scope and objectives of your research, and help you ‘fill in the gap’ where needed to gain the skills necessary to bring complete your research. 

For a complete list of supervisors, see Academic Staff section below.