It is not always as easy to be understood by digital tools. That is the experience of Frida Pernille Mikkelsen in Norges Mållag. She has to make several rounds before the iPhone assistant Siri finally understands that she is searching for the Bykle Labor Party and not the Sorry Party. Google’s transcription tool works even worse with the Førde dialect. Frida dictates into the automatic writing program: – Minutes from the information board. Frida, Vemund and Erik were at the meeting. First point of the meeting. I have to prepare some posts on the situation for main target students. And then it’s the king’s birthday, so I should have prepared a congratulations for that. But the text does not turn out as it should. It will be a porridge. – This report is very effective for me, laughs Mikkelsen. All dialects are important If everyone is to be able to use speech technology, it must understand all dialects, the Technology Council believes. Your expert committee recently submitted a report on this. – Grandma in Nord-Trøndelag is not used to speaking normally at all, says committee member Line Adde, product manager at Aidn. Adde has a PhD in speech recognition and works with digital solutions for the healthcare system. Knut Kvale, senior researcher at Telenor Photo: Telenor – Norwegian speech recognition today works best on Eastern dialects and slightly better on men than on women. It works very badly on children, says another committee member, senior researcher at Telenor Knut Kvale. People with Westland dialect often struggle to be understood by artificial intelligence, but Kvale also knows for himself how difficult it can be to be understood with Valdres dialect. It has to do with volume training, he explains. – If we had had unlimited voice recordings in the Valdres dialect, then the speech recognizer would have been much better at transcribing my dialect. The market is full of voice-controlled smart speakers. Photo: Tonje Grimstad / news Important tools The committee’s members also agree that such tools are becoming increasingly important. – Those who have obvious disabilities, obvious dyslexia, but also people who have a job where you write a lot. Because you actually speak 3–4 times faster than what you write, says Adde. Line Adde, product manager at Aidn Photo: Tonje Grimstad / news Today, for example, doctors can dictate the text into patient records. It’s especially good if your hands are busy with something else, says Adde. – In the operating theatre, for example. In the future, you will be able to control much more with your voice, says Adde. – Vacuum cleaners and ovens. Turn off the heat and light. Raise curtains. All such things are eventually voted on. Smart house then. Several robot vacuum cleaners can be controlled by voice using Google Assistant. Photo: Tonje Grimstad / news Needs a lot of data Speech technology is based on machines learning to recognize patterns. The Language Bank at the National Library works to collect audio and transcripts of Norwegians with different dialects, in order to offer this to those who develop speech technology. They can be researchers both at universities and in private companies. The National Library has its own investment in speech technology, through the Language Bank. Photo: Tonje Grimstad / news But arranging sound and prints is expensive, and large quantities are needed. This is according to language technologist Per Erik Solberg, who was also on the committee. – You need children. You need the elderly. We could use more everyday speech. Something we have a lot of, for example, is speeches from the Storting. It’s easy to get hold of, he says. The advantage of speeches from the Storting is that you know the age and dialect origin of everyone who speaks. In addition to the fact that the sound is not copyrighted. Asking the authorities to take action The authorities must get involved so that everyone can communicate with voice technology, the expert committee writes in the report. – I absolutely think so, says Adde. Per Erik Solberg, language technologist at Språkbanken Photo: Tonje Grimstad / news – One of the initiatives proposed in that report is something they have done in Finland. Which is to have an app or an online solution where people can freely record what they are talking about and submit it to a database, says Solberg. This in turn could be arranged for developers via the Language Bank. Culture and Equality Minister Anette Trettebergstuen agrees that it is very important that everyone can speak to digital aids in their own dialect. – Yes, it is important to say that language technology in Norwegian definitely does not come by itself. We must actively facilitate that, as a relatively small language area like Norway is, says Trettebergstuen. Minister for Culture and Equality Anette Trettebergstuen Photo: Ole Berg-Rusten / NTB She is also concerned that people who read audio should be able to have it transcribed in both Bokmål and Nynorsk. – Many people in our country have Nynorsk as their first written language, and they should have the same opportunities as Bokmål users to communicate digitally as well. The Minister of Culture believes that the authorities have already done a lot by establishing the Language Bank at the National Library. – But we are of course keen to consider all proposals that come in. Google is concerned about this. Google has some of the most widely used digital tools. Digital assistants from Google on display in the store. Photo: Tonje Grimstad / news Both speech recognition on Google Home, transcription tools and automatic translation of text. Head of Communications in Google Norway Sondre Ronander is not aware that their tool has any particular problems with Norwegian dialects. At the same time, he too, who speaks Eastern, may experience not being understood. – All our tools can always be improved. Of course. It is also the reason why, when you give a command to the assistant, it sometimes asks: Was this what you were actually wondering about? Sondre Ronander, Head of Communications at Google Norway Photo: Mari Reisjå / news Ronander says some languages, such as English, have a lot more material for the machines to train on. But he believes that Norwegian dialects are in principle no more challenging for the machines than dialects in other languages. – What is important is that people should be able to use their own goal, also as society becomes increasingly digitized, says the communications manager. – It is something you should be able to expect, and we are not there yet. Across the board, but that’s where we’re working towards getting to. Bruker Norge Rundt news contributes to the development of voice technology through collaboration with, among others, the Language Bank and NTNU. – We have made 116,000 programs available, with a wide range of dialects, says Egil Ljøstad in news. Egil Ljøstad, responsible for radio technology in the technology department at news Photo: Tonje Grimstad / news news can only share programs where it owns all the content. So far, they have shared around Norway and news broadcasts. They also plan to share more children’s programs so that the machine can be trained on children’s voices. news wants people with poor hearing to be able to receive an offer to print podcasts, for example, in the future, – We have over a million among our audience who have difficulties hearing, says Ljøstad. – We share the latest programmes, so the language is the way we speak today. Program leader Synnøve Sundby Fallmyr would like to show off her childhood alpine resort, but the weather is bad and there is little snow. Do they even get to ski?
ttn-69
– Must be able to speak dialect to digital assistants – news Norway – Overview of news from different parts of the country
