Macsen gets more talkative

Artificial Intelligence (AI) is getting more intelligent by the day. By now many people in Wales have devices such as Alexa, Siri, and Google Now that can answer spoken questions about the weather, news, and other useful facts. They can even respond to a voice asking them to turn on the light, switch on the power, or similar skills. Unfortunately, these only work if spoken to in English, but we are a step nearer getting a similar system in Welsh with our own digital personal assistant Macsen, due to work done at Bangor University, aided by a grant from the Welsh Government. To begin with Macsen could only respond to simple commands, but now it can answer questions, and go to the most popular articles in the Welsh Wikipedia to find relevant information. It can read out the first paragraph of an article, or the news headings, using a Welsh synthetic voice.    

An emerging small community of dedicated developers and hobbyists are beginning to code with Macsen. They offer new ideas and additional skills to the system. One of these is to make a bespoke ‘body’ for it, rather than cobbling together a mike and loudspeaker hooked up to a Raspberry Pi, as is done at present. Developments such as these will help not only Welsh speakers, but also the Welsh economy, as small local companies will be able to use the new technologies to understand apps with Welsh speech interfaces.  

Another advancement stemming from Macsen has been a transcription program which types on the screen words spoken by an individual.  Dewi Bryn Jones, the program’s lead developer at Canolfan Bedwyr's Language Technologies Unit explained:

“In reality the computer types what it thinks a person has said every time he or she gives a spoken command to Macsen. Showing the words on a screen is an extension of that. But the transcriber’s vocabulary is much larger than that of Macsen, and will improve in training. At present it works better with some specific voices, and those whose voices have been used to teach it.”

Eluned Morgan, Minister for the Welsh Language and Lifelong Learning said: “We live in a world where technology is developing very fast. It is essential that Welsh keeps up with the latest developments if we are to make sure that we reach the target of a million Welsh speakers by 2050. That is why I am glad to be supporting developments like Macsen.”

The code and resources to do this are available free of charge on the web, through the Welsh National Language Technologies Portal (http://techiaith.cymru). Any developer, company or hobbyist can use these resources to create practical products, and they are also suitable for use in Welsh coding clubs. 

Publication date: 10 April 2018