Voice-activated digital assistants that speak Scottish Gaelic could be one step closer thanks to a hi-tech advance by University experts.
A team of linguists and Artificial Intelligence specialists has developed software that can listen to spoken Gaelic and print it out as written text. Now researchers hope to upgrade the technology so it not only prints what it hears, but responds verbally too – just like voice assistants Siri, Alexa or Google.
The speech recognition system can already provide subtitles for online video content. It can also help those who are learning the language and support Gaelic-medium schoolchildren with dyslexia, the researchers say.
The team – led by Edinburgh academics William Lamb and Beatrice Alex – collected millions of spoken and written Gaelic words and trained a computer system to recognise how they were related. Researchers did so using a neural network – AI that enables computers to process language in the same way that humans do.
The software was developed in tandem with Mark Sinclair, of University of Edinburgh spin-out company Quorate Technology Ltd. Also involved were the University of the Highlands and Islands and the Tobar an Dualchais/Kist o Riches project – a unique online record of Scotland’s rich oral heritage.
The team has begun working with Tobar an Dualchais to transcribe interviews with Gaelic speakers that include precious elements of oral history and traditional storytelling.
Project leader Dr Lamb, of the University’s School of Literatures, Languages and Cultures, said: “Ensuring that Gaelic has a place in the modern technological landscape is key for its survival. By enlisting the support and expertise of the Gaelic community, and giving back to them in this way, we hope to demonstrate that any minority language can thrive in the digital age.”
Words are all I have
Researchers say the challenges of setting up automated speech recognition for Gaelic are immense compared with a similar system for English.
Anyone creating an English ASR system has a wealth of publicly available data to draw from, as well as technologies tailor made to assist with the process. State of the art English systems today are trained on one million hours of audio, which is more speech than any individual will hear in a lifetime.
Good quality, transcribed speech data is generally not so easy to come by in minority languages. The Gaelic initiative is more or less starting from scratch with just 65 hours of audio, but a deal with Gaelic broadcasting organisation MG Alba will boost this significantly.
In the world of machine learning, the more data you have, the better your system will be. So, with less data available for these languages, it’s harder to get a better system up and running.
Researchers can, however, use a number of mediating methods to boost the performance of a low-resource system. The key, according to Dr Lamb, is finding what works best for the dataset.
There are other handicaps. Despite the relatively small area in which Scottish Gaelic is spoken today, there is dense dialectal diversity and no accepted standard.
“English models are geared towards middle-class Home Counties in the UK, or standard east coast dialects in the US, but that approach doesn’t work with Gaelic,” says Dr Lamb. “The most common dialect today, Lewis Gaelic, is markedly different from the dialects of neighbouring Harris and North Uist. So, despite the fact that we are catering to a small fraction of the speakers that you would have with English, our challenges are, in many ways, much more pronounced.”
Gaelic for all
Despite the obstacles, the team is hopeful that the emerging technology can help widen access to Gaelic for anyone who wants to learn it.
For proof that speech technology can broaden Gaelic’s appeal, look no further than project team member Lucy Evans. The early career researcher grew up bilingually in Switzerland, speaking English and Italian, before moving to the UK for secondary school.
“I just find the ability of computers to pick up on all the complexities of language so interesting,” says Lucy, who completed an MSc in Speech and Language Processing at the University of Edinburgh. “Luckily, for Scottish Gaelic, there’s already a digital dictionary, which essentially maps any written Gaelic word to its phonetic pronunciation. As long as you have the necessary resources, it’s the computer that does the language learning.”
The project is backed by the Data-Driven Innovation (DDI) initiative, which is led by the University of Edinburgh and Heriot-Watt University and a key part of the Edinburgh and South East Scotland City Region Deal. DDI supports the Region in its bid to become the data capital of Europe and its Building Back Better fund awards grants of up to £500,000 to research that builds economic and social recovery from Covid-19.
The speech technology project also received generous support from Soillse, the National Research Network for the Maintenance and Revitalisation of Gaelic Language and Culture.
Picture credit: lucentius/Getty; Daisy-Daisy/Getty