Words matter - Language technology in the Norwegian Open AI Lab
The term language technology refers to theoretical and applied informatics aiming to enable computers to "make sense" of the human language. It includes a variety of methods, including artificial intelligence methods.
The interest in language technology, and particularly the AI subfield of natural language processing (NLP), has increased rapidly in recent times, as a consequence of its many potential application areas and the technological advances that have been made in the field. The potential for language technology is vast in business, industry and the public sector. Indeed as consumers, we've all become used to interacting with chatbots in customer service, and NLP-powered software is increasingly being used while dealing with large quantities of text, for instance legal documents, newspaper articles and patient journals.
Still a job to be done for Scandinavian languages
However, the most impressive breakthroughs within NLP have been made in the English language. For Norwegian and the Scandinavian languages, there is still a job to be done. Luckily, our Norwegian Open AI Lab partners and researchers are on the task. Ole Jakob Mengshoel, professor at the Department of Computer science, NTNU, is one of the researchers working with improving language technology for Scandinavian languages.
Language technology is a fascinating field in that humans deal with most aspects of speech and text (in their own language) with ease, yet it has proven challenging to replicate this with technology. Tasks such as automatic speech recognition and natural language processing have been worked on by AI researchers for half a century. Despite the great progress that have been made during the last 10-15 years, researchers have been unable to find solutions to important challenges", says Professor Mengshoel.
NAIL researchers are currently working on different language technology projects. Here are some of them:
Improving speech-to-text transcription for Norwegian
Transcription from speech to text can be challenging, especially when it comes to real-life conversations, where background noise, interruptions, laughter and spontaneous turn-taking may disturb the language flow. For Norwegian, the many different dialects complicate this further. Researchers from NTNU and Telenor aim to tackle this challenge by developing automatic speech recognition (ASR) technology for Norwegian, as part of the project SCRIBE, funded by the Norwegian Research Council. The Norwegian National Library and NRK are also partners in the project, and contribute with data and expertise.
New methods and tools will be designed to improve transcriptions of real-time conversations. For Telenor, this could mean a pulse check on customer sentiment, and for the hearing impaired, it could enhance their experience with live events or when watching tv and listening to radio or podcasts.
.. and for Swedish too
However, NAIL researchers are not only focusing on developing ASR for Norwegian. In the project NORDTRANS, funded by the EEA Grants, similar work will be done for both Norwegian and Swedish language. Moreover, this project involved Czech partners from the Technical University of Liberec and the company NEWTON Technologies, which develops ASR applications.
The ASR technology to be developed will operate with high accuracy in various applications, such as broadcasting in different types of media, as well as transcription of public speeches. The technology will be incorporated into already existing multilingual speech processing solutions from NEWTON. Their ambition is that this will open a new market and cooperation opportunities. This will also bring new and improved services based on ASR closer to Scandinavians.
illustration language project
NAIL projects within language technology
- TEFLON - Technology-enhanced foreign and second-language learning of Nordic languages | Giampiero Salvi & Torbjørn Svendsen
- NordTrans - Technology for automatic speech transcription in selected Nordic languages | Giampiero Salvi & Torbjørn Svendsen
- SCRIBE - Machine transcription of Norwegian conversational speech | Giampiero Salvi, Torbjørn Svendsen & Ole Jakob Mengshoel
- MEGAS - Metadata generation of auto-transcribed text | Ole Jakob Mengshoel
- e-LADDA - Early language development in the digital age | Giampiero Salvi & Torbjørn Svendsen
language tech 2
Applying language technology in language learning
Language learning is another area in which language technology can be applied. In the project TEFLON, funded by Nordforsk, partners from three Nordic countries are coming together to develop new techniques of teaching languages that are suitable for digital application.
This project targets immigrant children arriving in the Nordic and who are in the process of learning a new Scandinavian language. The demand for learning Nordic languages is high, yet acquiring a new language is challenging for many learners. To speed up individual learning at the same time as avoiding high costs of tutoring for the Nordic societies, digital solutions applying new techniques of language teaching are needed.
Furthermore, since gaming is often a hit among children,a digital language-learning game is at the center of the project. The game will be available in Finnish, Norwegian and Swedish, and will include ASR used to assess children's utterances and provide feedback to reinforce learning. This will require advanced speech technology that not only can assess foreign speech, but also children's speech.
Similar work is also being done in Early Language Development in the Digital Age, an EU-funded PhD project including 14 PhD candidates from different universities across Europe. In one of these projects, NTNU PhD students study methods to improve the performance of speech recognition of children's voices without having to use large datasets of children's voices.
Metadata in the spotlight
Yet another area for language technology research is related to information extraction from audio sources, such as broadcasting archives. For large archives, having good quality metadata is crucial in order to increase the availability, access to and utilization of the data that is stored. The NAIL-funded MEGAS project has been established in conjunction with the SCRIBE project (referred to above), with an ambition to improve the automated generation of metadata from auto-transcribed speech.
Text generated by auto-transcription has different characteristics than normal text, which need to be taken into account when the ambition is to generate high-quality metadata. One case in the MEGAS project is based on an NRK dataset, containing captioned TV programs with Norwegian speech, in which the desired outcome for NRK is a technology foundation for more advanced searching features for their TV and radio program archives. NTNU researchers, master's students and NRK staff will collaborate to try to make this a reality.
As one can see from the variation of the projects mentioned here, language technology has many potential application areas and can contribute positively in several domains. If you have an interesting language technology case for Norwegian or would like to learn more about our ongoing projects, feel free to contact us.