Council slashes translation costs with machine learning

Swindon Council in the UK deployed a machine learning solution which cut content translation costs from £160 (US$203) per document to just 7p and reduced turnaround time from weeks to minutes.

The code for the system has been made available as open source to benefit other local governments.

The initiative came out of a request from the council’s Paediatric Therapy Service. Swindon has a population of 230,000 and around 100 languages are spoken across the borough. In 2021, the Paediatric Therapy Service had around 400 documents translated, costing on average £160 each and taking up to 16 days. This included internal preparation and sending files to an external translation agency.

Swindon’s Emerging Technologies team undertook a market assessment for solutions to make translating content faster and cheaper, including local bilingual speakers in the selection process. They eventually selected Amazon Web Services (AWS).

“[AWS] had some technology, and this was a good use case that they hadn’t considered for it,” said Chief Digital Officer Philip Murkin.

The team worked with AWS to develop the machine learning tool for the council’s requirements, based on the Amazon Translate neural machine translation service. They collaborated to reduce the process to three clicks and designed a user interface. The system needed to be fast, easy to use and resilient, as well as secure to protect child information.

With no upfront cost for Swindon other than staff time, the council reports that the tool has reduced translation spending from £64,000 a year to £27 in the Paediatric Therapy Service. Documents are available in an average of 14 minutes.

The document translation tool is now also being used in other council departments, Murkin told Cities Today.

A year since it was implemented, he says that as well as saving time and money, the solution is meeting accuracy requirements and the council hasn’t received any clarification questions related to the translations.

Sharing innovation

Other councils can now also benefit from the work done by Swindon.

“We’re very passionate about open source and sharing solutions across the public sector,” Murkin said.

The council stipulated in its contract with AWS that in return for co-developing the product, it should be provided to other customers “at cost”.  It is available as open source via GitHub and AWS Industry Solutions, with translation on a pay-as-you-go basis.

According to AWS, the solution is also being rolled out in other UK councils, including Edinburgh, as well as in countries such as France and Spain.

“When Swindon showed us what they built, it became immediately apparent this was game-changing for organisations like Swindon – but in fact not just for Swindon, for local authorities, councils and municipalities around the world,” said Wayne Soutter, Principal Industry Solution Manager, AWS.

“We helped them make the solution more generic and also built in security and user management.”

Emerging technology

Swindon’s Emerging Technologies team was set up in 2019 to identify opportunities that benefit multiple council services and test technology. The team will also look at how it can use a similar machine learning solution to meet easy read requirements in adult social care.

Other projects have included using drones for roof inspections, deploying AI to detect potholes and fly-tipping, and implementing robotic process automation for Free School Meals processes.

“There are lots of opportunities for innovation out there and lots of use cases,” said Murkin. “If you partner with the right organisations and you speak to some of your peer local authorities, I think we can do a lot together to really bring benefits to customers. And if we share more of those use cases between us, then we’ll all be in a better place.”