Every year, large volumes of operational audio are reviewed across government systems. Some recordings are in foreign languages, making them difficult to analyse quickly and accurately.
Last month, we shared our work exploring how artificial intelligence (AI) could help transcribe and translate foreign-language audio at the Government Operational Research Service (GORS) conference in London.
GORS brings together analysts from across government to share ideas, research and real-world applications of operational research and data science. This year’s conference, themed “Intelligent Analysis”,explored how analytical methods and emerging technologies like AI can support better decision-making across the public sector. Around 300 delegates attended each day, making it a great opportunity to learn from colleagues across government and share our work.
Evaluating AI for transcription and translation
At the conference, we presented a project exploring how AI tools could support transcription and translation of foreign-language audio data.
Processing multilingual audio can be time-consuming and difficult to scale using traditional approaches. Human transcription and translation provide high-quality results but can be slow and resource-intensive with large volumes of material.
Our project explores whether AI tools can help improve parts of this process, enabling analysts to review information more quickly.
However, this is not a straightforward translation task. Operational audio can contain informal language, slang, overlapping speech and challenging audio conditions that mainstream tools are not always designed to handle.
To understand how well current tools perform, we benchmarked models from several leading AI providers and compared outputs against human translations to assess accuracy, consistency and usefulness.
Using AI responsibly
Using AI in sensitive analytical contexts requires careful thought.
During the talk, we shared how we applied the MoJ AI and Data Science Ethics Framework, which helps teams assess risks and ensure that AI and data science projects are developed and used responsibly.
This includes considering issues such as fairness, transparency and appropriate human oversight when deploying new technologies.
MoJ work across the conference
It was great to see other Ministry of Justice colleagues presenting their work, highlighting the wide range of analysis taking place across the justice system.
One session explored how a new Release Outcome Measure is providing earlier insights into outcomes after people leave prison. Another showed how microsimulation modelling is being used to assess the impact of reforms designed to address the Crown Court backlog.
What we learned
Context matters. Tools that perform well on standard benchmarks do not always perform the same way when applied to complex operational data. Testing them against real-world examples is essential.
AI works best alongside human expertise. Automated transcription and translation can help analysts work more efficiently, but human judgement remains critical for interpreting meaning and nuance.
Responsible use must be built in from the start. Applying the MoJ AI and Data Science Ethics Framework helped ensure that questions of fairness, transparency and oversight were considered throughout the project.
Reflections from presenting
We delivered our talk to an audience of more than 150 analysts and policymakers, and it was encouraging to see strong interest in how AI could address complex operational challenges.
Events like the GORS conference provide a valuable opportunity to share ideas, learn from colleagues across government and showcase how analytical work can support better decisions and improve public services.
Get involved
If you are interested in how operational research, data science and AI are being used across government, the GORS community offers a great opportunity to connect with analysts tackling complex challenges.
seen at 15:08, 24 March in Justice Digital, Data and Science .