Language Models for Swedish Authorities
The project is coordinated by RISE (Research Institutes of Sweden), which here describes it as follows: “The project provides tools and resources to enable Swedish public authorities to develop and integrate cutting-edge language technology solutions into their current and future services. The development of large-scale Swedish language models specially designed for use in the public sector and the techniques for using them will allow for new types of language technology applications that will be in a position to revolutionise the use of language technology in the Swedish public sector. The data, infrastructure and frameworks that are developed during the project will also enable new types of representation learning to be created and ensure that Swedish language technology is at the forefront of development”.
Part of the work on the project will involve determining exactly which language models should be developed. The four use cases that the project will initially focus on are machine understanding and conversational AI, semantic textual similarity, entity tagging (named entity recognition) and text classification.
In addition to RISE and AI Innovation of Sweden, the other partners in the project are Luleå University of Technology, Peltarion, the Swedish Public Employment Service, the Swedish Agency for Economic and Regional Growth and the Swedish Tax Agency.
AI Innovation of Sweden is responsible for coordinating and validating the data used in the project and for distributing the results and the knowledge to the reference group, external stakeholders and interested parties. One way of doing this will be to organise workshops and events (such as hackathons) and to cooperate and share knowledge, experience and data with the other two NLP projects and with other data owners.
By using AI to make communication more automated, it will be possible for society to significantly reduce the use of resources. Public authorities, for example, can have help with compiling and summarising reports and cut the normally long queuing times. When authorities need to manage large amounts of documentation, NLP will allow them to link together different text documents based on their content which will be very valuable, for example for matching applicants to job adverts at the Swedish Public Employment Service. These are just some examples of how public bodies can use the language models developed during the course of the project as tools that will make it easier for them to sift through, categorise and find the right information.
As the project is still to some extent in the initial phase, the most apparant milestones are currently the collection of data and the development and implementation of models.
WP1: Data and evaluation (AI.se 70% of WP1 hours)
WP2: Algorithms and architecture
WP5: Distribution of results and coordination (AI.se 90% of WP5 hours)