Real-Time Translation

This information was sourced from "Real-Time Translation" by AI.GOV.UK, https://ai.gov.uk/knowledge-hub/tools/real-time-translation/?directTo=library. Licensed under Open Government Licence v3.0.

The challenge

Border Force and Immigration Enforcement officers frequently need to communicate with people who speak little or no English. Currently, they rely on human translation services to support these discussions, which can be expensive. The question is whether automated transcription and translation can be effectively and ethically integrated into these operations while maintaining accuracy and reliability.

The solution

The solution provides secure, scalable and multi-language translation capabilities. Using advanced speech-to-text and machine translation models, the tool can handle over 100 languages and supports two-way translation, enabling near real-time communication between officers and service users. Audio is captured via a secure interface, processed through AWS AI services for transcription and translation, and stored in compliance with Home Office governance. This approach will deliver faster, more accurate communication between officers and service users at lower costs by alleviating the need to use human translation services. It will also reduce reliance on handwritten notes, improve operational efficiency, enhance safeguarding, and provides secure, auditable transcripts that can unlock intelligence insights and future data-driven opportunities.

The results

The tool is still in the early stages of user testing, but initial feedback has been very positive. Users report that the tool is intuitive and will significantly improve their interactions with service users. A standout feature is the auto-detect language function, which is particularly valuable in situations where the officer does not know the service user's language. Early feedback also highlights additional benefits, such as access to richer insights and data that were previously unavailable, supporting better analytics and intelligence activities. As we progress further through user testing, we will provide further insights on the impact of RTT.

Learnings / insights

• Start small - piloting in controlled environments before full rollout helps identify limitations / issues.
• Good test data is hard to find - especially data that's the right size, quality, and publicly available.
• Technology moves fast - AI models evolve quickly, and documentation doesn't always keep up.
• Different operational contexts require different accuracy thresholds and tend to encounter different languages.
• Supporting 100+ languages is valuable, but performance and latency vary by language.