ICS.AI is developing a framework for linguistically inclusive conversational AI with the University of Sheffield, and applying the work in live public service settings across the UK.
The framework is designed to help conversational AI systems understand the range of regional accents, dialects and everyday phrasing used by people accessing public services. It combines sociolinguistic research, operational development and existing council deployments to test how these systems perform beyond standard accuracy measures.
Work is already under way with Derby City Council, Renfrewshire Council and Bristol City Council. In those settings, ICS.AI is designing its systems to respond to natural variation in speech so residents do not need to change how they speak when using digital services.
The project builds on a wider partnership between ICS.AI and the University of Sheffield. Earlier research commissioned with the university examined how well digital public services understand speakers with different regional accents and dialects, an issue that has drawn growing attention as conversational AI becomes more common in local government, higher education, healthcare and social care.
Live testing
Rather than relying only on lab-based benchmarks, the latest phase looks at how AI systems behave in frontline services, where people use informal language and local speech patterns. The aim is to shape how conversational systems are designed, evaluated and deployed in public sector settings.
The UK offers a useful test bed because of its dense linguistic variation. ICS.AI argues that lessons from Britain could apply more broadly across Europe, where language differences often exist not only between countries but also within regions, communities and local service areas.
For public bodies, the issue goes beyond technical performance. If speech-based systems struggle with accents or dialects, residents may face uneven access to information and support, particularly in services where speed and clarity matter.
The concern is becoming more pressing as councils and other public organisations increase their use of automated tools to handle enquiries, direct people to services and manage routine interactions. In that context, how well systems understand different speech patterns can affect whether a service feels accessible and fair.
Dr Crispin Bloomfield set out the company's view of that challenge.
"Public services exist to support everyone, and citizens should not have to change the way they speak in order to access essential services. As conversational AI becomes embedded into everyday interactions, there is a responsibility to ensure these systems work fairly, effectively, and naturally across the full diversity of spoken English in the UK. This work is about moving beyond theory and building practical, responsible AI frameworks that reflect how people actually communicate," said Dr Crispin Bloomfield, Chief Education Solutions Officer, ICS.AI.
Research link
The University of Sheffield's involvement brings dialectology and sociolinguistics into a field often dominated by general performance measures. The collaboration is intended to connect academic evidence with data from public service deployments, giving both sides a clearer picture of how systems manage variation in spoken English.
Dr Chris Montgomery described the value of combining research with live operational evidence.
"By grounding their operational AI in rigorous sociolinguistic research, ICS.AI are innovating to extend their systems to further adapt to how people actually speak. Our collaboration represents a rare bridge between academic evidence and live deployment data, allowing us to move beyond aggregate benchmarks. Ultimately, our joint goal is to measure success by how equitably our technology serves diverse voices, ensuring no regional accent or local dialect receives an uneven experience of AI enabled public services," said Dr Chris Montgomery.
ICS.AI's investment is focused on practical standards and operating approaches for dialect-aware AI in public services. Its current work with councils is being used to refine those methods in real settings where residents bring a wide range of speech patterns to everyday interactions.