HALO: Artificial Intelligence for Dynamic, Context-Responsive AAC

Abstract

Background: Adults with acquired or progressive neurological conditions, including aphasia, Parkinson’s disease, amyotrophic lateral sclerosis, and other neurodegenerative disorders, may experience complex communication needs involving speech, language, motor access, cognition, or combinations of these factors. These difficulties can reduce communication rate, compromise access reliability, restrict participation, and affect personal and vocal identity. Conventional augmentative and alternative communication (AAC) systems provide essential support, but vocabulary often needs to be anticipated and may not adapt quickly to changing environments, topics, or communication partners. Generative artificial intelligence may reduce message-generation effort and support more personalised and contextually relevant communication, while also introducing risks related to authorship, accuracy, privacy, consent, data use, cognitive load, and user control.

Focus of the presentation: This session presents the clinical and technical implementation of HaloNeuroAI, a personalised artificial-intelligence-supported AAC system designed for people with complex communication needs. The system combines context-aware language generation, personalised communication profiles, synthetic or voice-cloned speech output, multimodal access options, dynamic communication boards, and agent-based functions that support communication and selected digital actions. The presentation will describe how these components can be configured around the person’s language, relationships, routines, preferred expressions, communication contexts, motor abilities, and changing access needs.

A central focus will be the use of AI-generated dynamic communication boards. These boards can be created or adapted according to the user’s context, interests, goals, and vocabulary requirements. The user can generate, select, and modify their own boards, independently or with support, rather than relying exclusively on boards prepared by a clinician or communication partner. Boards may include text, symbols, photographs, and animated GIFs. Movement-based visual content may support comprehension, action representation, sequencing, engagement, or communication about dynamic events.

The session will also introduce Ambient Semantic Awareness (ASA), a context-aware feature that uses a photograph of the surrounding environment, object, activity, or situation as a starting point for communication. From the image, the user may select what they want to say, or the system may generate a context-specific communication board. This may reduce the need to predict all vocabulary in advance and support communication in unfamiliar, rapidly changing, or less structured everyday situations, while preserving user choice and confirmation.

Practice-based examples from implementation with adults with aphasia, Parkinson’s disease, amyotrophic lateral sclerosis, and other neurodegenerative conditions in Portugal will illustrate potential uses in everyday conversation, asynchronous voice messaging, communication outside the home, participation in meaningful activities, and interaction with family members and professionals. Examples will show how features may be adapted to different language, speech, cognitive, sensory, and motor profiles. The role of the speech and language therapist will be examined across assessment, feature matching, personalisation, access-method selection, communication-partner training, monitoring, and adaptation, with attention to functional communication goals and participation outcomes.

Clinical safeguards: Generated messages should remain attributable to the user, reviewable, editable where appropriate, and subject to explicit confirmation whenever the system could alter, expand, or misrepresent intent. Personalisation should not result in the system speaking for the person or producing language that exceeds their preferences, communicative purpose, or decision-making authority. Dynamic board generation and image-based processing should support choice rather than replace it. Voice cloning requires specific informed consent, secure data handling, transparent information about future use, and clear withdrawal procedures. Photographs and contextual data also require careful governance, particularly when they include other people, private environments, or sensitive information.

Participants will be invited to consider practical criteria for evaluating usability, communication efficiency, authenticity, participation, cognitive demands, access reliability, learnability, partner support, and user agency. The presentation will also address limitations, system errors, unsupported outputs, and contexts in which conventional AAC strategies may remain more appropriate.

Conclusion: Generative artificial intelligence can extend AAC beyond faster text production and static vocabulary sets. AI-generated dynamic boards, animated visual content, context-aware image processing, personalised language support, and voice preservation may contribute to more responsive communication, participation, continuity of identity, and digital autonomy. Their clinical value depends on person-centred implementation, transparent system behaviour, preserved authorship, accessible confirmation mechanisms, data protection, and meaningful outcome measurement. HaloNeuroAI will be presented as an implementation case through which the opportunities, limitations, and governance requirements of personalised generative artificial intelligence in AAC can be critically examined.

Use of artificial intelligence in abstract preparation: Generative artificial intelligence was used to support the initial structuring and language editing of this abstract. The authors provided the substantive clinical and technical content, critically revised all sections, verified the factual statements, and take full responsibility for the final submission. Artificial intelligence was not used to generate, analyse, or interpret research data.