Nurses at Inverurie Hospital are testing an AI note-taking tool while teaching it the Doric dialect spoken across the North East of Scotland. The work is taking place on busy wards where accurate clinical notes matter. Staff say the effort aims to save time and capture patient voices more clearly.
The move reflects a wider shift in healthcare toward speech-to-text systems. It also spotlights a long-standing problem in voice technology: understanding regional speech. Doric, a form of Scots, presents unique sounds and words that standard systems often miss. That can affect accuracy, safety, and trust.
Why Doric Matters in Care
Doric is part of daily life in Aberdeenshire communities. Patients often use Doric at the bedside. When tools fail to understand them, clinicians must repeat questions or switch to typing. That slows care and risks losing detail.
Language barriers are not limited to foreign languages. Dialects can lead to misheard medication names, symptoms, or dates. For nurses, even small errors in notes can create extra work later. A system that recognizes local speech could reduce friction during handovers and discharge planning.
What Staff Are Doing
Ward teams are introducing the tool during routine assessments and follow-up conversations. They are correcting transcripts in real time. The system then learns from those edits and builds a better model of Doric vocabulary and pronunciation.
“Nurses at Inverurie Hospital are using an AI tool to take notes while helping it understand Doric.”
Clinical leads describe a phased approach. Early use focuses on low-risk interactions, such as summarizing background history or capturing social details. More sensitive tasks remain under close human review. The goal is to confirm accuracy before expanding use.
Potential Gains and Practical Limits
Advocates expect the tool to reduce time spent on paperwork. That could free staff for direct care. It may also improve the completeness of records if more patient words make it into notes.
But voice systems struggle in noisy wards. Masks, alarms, and accents can reduce accuracy. Staff report that consistent microphone use and clear prompts help. They also note that correction features must be quick, or the tool adds steps instead of removing them.
Voices From the Ward and Community
Nurses say patients respond well when they see local language reflected in their records. Community advocates for Scots languages view the effort as recognition of cultural identity. They argue that systems should work for people as they speak, not the other way around.
Some clinicians caution against overreliance. They remind colleagues that notes are legal documents. Every line must be checked. A speech tool can help, but it cannot replace clinical judgment.
Privacy, Safety, and Governance
Digital safety is central to any AI deployment in healthcare. Data protection experts point to several guardrails that should be in place:
- Clear rules on where audio and text are stored.
- Encryption in transit and at rest.
- Strict access controls and audit logs.
- Options for patients to opt out when appropriate.
- Transparent policies on model training and data retention.
Clinical safety checks are equally important. Hospitals typically run accuracy audits, bias assessments, and incident reporting for digital tools. For dialect support, teams should track error rates by key terms, such as medication names and allergies.
What Success Could Look Like
Reliable recognition of Doric could shorten note-writing time and cut duplicate questions. More precise documentation may reduce missed details during transfers. Training the model on local speech could also help visiting staff understand colloquialisms captured in notes.
Hospitals elsewhere are watching dialect pilots closely. If systems can learn Doric, they may adapt to other regional forms across Scotland and the UK. That would widen access and improve equity in digital care.
Measuring Impact
Experts recommend simple metrics to judge progress. These include minutes saved per shift, correction rates, and user satisfaction. Patient feedback also matters. If people feel heard, they are more likely to engage with care plans.
Longer term, leaders will look for links to outcomes. Better notes should support safer prescribing, clearer referrals, and fewer readmissions. Any gains must be proven and maintained over time.
The Inverurie effort shows a practical path for bringing AI to the bedside without losing the human touch. Staff remain in control while the system learns the language of the community. The next steps include broader testing, stronger privacy controls, and open reporting of results. If the tool keeps improving, it could set a model for dialect-aware care across the country.
