A new study led by the University of Sheffield could help ensure people are no longer misunderstood by automated phone systems due to their accent.
ICS.AI is a UK-based AI company aiming to transform public sector organisations. Their research explored how conversational AI handled different accents and dialects across the UK.
Dr Chris Montgomery, Senior Lecturer in Dialectology, said the system would be straightforward if such research were embedded into AI systems.
“People would be less likely to be misunderstood because of their accent or use of local expressions. That means fewer times having to repeat yourself and a smoother experience overall.”
He stressed that Sheffield City Council isn’t currently using ICS.AI systems and he isn’t aware of any plans for it to do so. The broader aim is to make sure public services work as efficiently as possible, regardless of if they have a Sheffield accent, a Scottish accent or any other regional variety.
Regarding Sheffield and South Yorkshire accents, Dr Montgomery believes there is “no clear evidence” that the accent is more difficult to interpret than others. This, instead, may be down to sound quality or how these systems are trained.
“AI systems can sometimes struggle with regional speech, but that depends on a lot of things.
“Companies usually measure this by looking at how often the system mishears words or misunderstands what someone is trying to do.”
Dr Montgomery will be publishing his findings from his research widely, sharing his insights with the broader community so that genuine progression can be made:
“My academic research is published widely, so the general findings about accents and languages are available to anyone. If I work directly with a particular company, there may be some commercial details that stay with them. But the wider principles, like how to test systems fairly across different accents, are relevant to the whole sector, not just one provider.”
While some improvements could happen quickly through testing and trialling, Dr Montgomery emphasised that AI isn’t a ‘one off fix’
“It’s an ongoing process. Systems improve over time as they learn from real-world use and as developers pay closer attention to linguistic diversity.”