Norwegian conditions is a broad and vague concept that can encompass very much. Some will apply generally for use in Norway, but much will apply specifically to health and care services in Norway. We will look more closely at the latter.
Based on acquired knowledge, adaptation to Norwegian conditions can be divided into five areas: languages in Norway, general knowledge in Norway, public administration in Norway, legislation in Norway, values and ethics in Norway. Similarly, adaptation to Norwegian health and care services can be divided into five areas: health professional language in Norway, health professional knowledge and practice in Norway, health administrative knowledge and practice in Norway, Norwegian health legislation, and values and ethics that apply in health and care services [96].

Languages in Norway
Bokmål and Nynorsk
Language models adapted to Norwegian conditions must be able to perform tasks in Norwegian. The Language Act provides several guidelines for public language use in Norway. According to Section 4 of the Language Act, Norwegian is the national main language in Norway, and Bokmål and Nynorsk are equal written languages in public bodies [97]. In addition, public bodies shall comply with the official orthographic rules for Bokmål and Nynorsk. Language models for use in public health and care services must therefore master Bokmål and Nynorsk within the established orthography. Language models should also be able to handle variation within Bokmål and Nynorsk.
The Language Council states in a linguistic survey from October 2024 that the language in ChatGPT did not maintain sufficiently high quality to be used by public administration [98]. Evidence suggests that language models are still not well enough adapted to Norwegian language.
Sami Languages
Sami languages, i.e., Lule Sami, South Sami, and North Sami, have status as indigenous languages. According to Section 5 of the Language Act, Sami languages and Norwegian language are equivalent [99].
The purpose of the Sami Act is to create conditions for the Sami ethnic group in Norway to secure and develop its language, culture, and community. It sets requirements for when public bodies must use Sami languages. At the same time, the law gives individuals linguistic rights in meetings with public bodies. Section 3-5 of the Sami Act gives persons extended right to use Sami in health and care institutions in the administrative area for Sami language [100][101][102].
Work should be done to develop and make available language models that function in Sami taking into account both the linguistic and cultural perspective.
Other Languages in Norway
Norway is a multilingual country, and we find a range of other languages in use in addition to Norwegian and Sami languages: the national minority languages, Scandinavian, Norwegian sign language, and newer minority languages.
North Sami, South Sami, Lule Sami, Kven, Romani, and Romanes are all defined as minority languages in Norway and thus protected by the minority language charter [103]. Speakers of these minority languages do not have their own concrete rights related to meetings with health and care services. As with Sami language users, efforts should be made to ensure that AI tools consider the cultural perspective in meetings between users from minority groups and society at large, which will be in line with the Patient and User Rights Act. In the health and care sector, this will largely apply to the newer minority languages.
Scandinavian languages can be used in patient records. According to Section 10 of the Patient Record Regulation, Danish and Swedish can be used to the extent that it is responsible [104].
Multilingual Language Models
Norwegian health and care services use an intricate mix of Norwegian and international medical terminology. This requires advanced embedding spaces that can handle multiple languages simultaneously.
An embedding space is a multidimensional mathematical space where words, sentences, images, or other data are represented as vectors. Similar concepts are placed near each other based on semantic or contextual similarity.
Modern language models implement this through shared vector spaces for medical terminology, while maintaining separate representations for Bokmål and Nynorsk with a common medical vocabulary as foundation. Extensive testing has shown that this architecture, especially with dedicated medical embedding levels, results in a significant increase in precision in cross-linguistic medical communication.
Source: https://link.springer.com/article/10.1007/s10462-025-11162-5
Section 6 of the Interpreting Act gives public bodies a duty to "use an interpreter if this is necessary to uphold legal safeguards or provide proper assistance and services. In the assessment of whether it is necessary to use an interpreter, weight shall be given to factors such as whether the parties to the conversation can communicate adequately without an interpreter, as well as the gravity and nature of the matter." If language models are to be used in such cases, for example for automatic interpretation, the Interpreter Act will probably apply. Language models must then be adapted to relevant languages.
Norwegian health professional language
If language models are to be used in health professional contexts, they should in addition to Norwegian master the Norwegian health professional language, including professional terminology. Professional terminology includes specialized professional terms, abbreviations, and jargon.
The professional language is characterized by many synonyms, where Norwegian, Greek, Latin, and hybrid (Latin terms adapted to Norwegian spelling) are often found in combination. In addition, professional terminology varies greatly between different professional groups and specialties.
Language models should therefore be able to both receive instructions and produce text with established and correct Norwegian terminology for the relevant area of use. It would, for example, be unfortunate if a language model automatically translates and creates completely new direct translations where there are well-established professional terms that have standing in Norwegian professional language. This has so far been a challenge when using AI tools for translation.
Dialects
If language models are to be used to process speech, for example speech-to-text, the language models should also master the dialects in use in Norway as well as other oral speech variations such as Scandinavian languages. The use of dialects has been a significant challenge with previous language technology. However, newer generative language models like NB-Whisper have shown promising results. This becomes particularly important for several types of tools, for example speech-to-summary (ambient scribe solutions) and oral language use in chatbots (conversational AI).
NB-Whisper
The National Library has adapted the language model Whisper from the company OpenAI to Norwegian. It is a language model that can convert speech to text, for example conversations and monologues.
NB-Whisper has been trained on large amounts of language resources from the National Library, including Norwegian speech corpus. This makes the language model understand Norwegian speech and Norwegian dialects far better than previous technology.
Like other generative models, NB-Whisper can also experience hallucinations, which is particularly important to be aware of in health and care services.
NB-Whisper is freely available for use, also in the health and care sector.
NB-Whisper is a good example of how an international language model can be adapted to Norwegian conditions. Systematic work with building good language resources at the foundation has facilitated that the public or industry can develop solutions for health and care services. Several technology suppliers now offer speech-to-text technology based on NB-Whisper.
NB-Whisper is made available by the National Library, which manages and further develops the model.
Plain language
The Language Act sets requirements for plain language, i.e., clear and correct language adapted to the target group. In addition, Section 3-5 of the Patient and User Rights Act is relevant: When information is to be given about a patient or user's health condition and health care, the information shall be adapted to the recipient's individual prerequisites, including culture and language background. Language models should therefore be able to differentiate language use and choice of words depending on who the users/readers are, whether they are citizens with different levels of health literacy or professionals.
Difference between languages is not always about different words and sentence structures. Language is also a cultural expression. Norms for good language use, such as politeness, style, and tone, vary between cultures and languages. An adapted language model should therefore also be able to express itself in a way that is in line with norms for good language use in Norway.
Health professional knowledge and practice in Norway
Health professional knowledge can be international and independent of countries and cultures. At the same time, much health professional knowledge is conditioned by specific cultural areas. For example, knowledge about diet and nutrition and associated advice can vary from country to country, and one often speaks of Nordic nutrition recommendations [105]. A language model adapted to Norwegian conditions must be able to build on Norwegian (or Nordic) knowledge and our health professional understanding of the world, as the nutrition example illustrates.
Language models used in health and care services should also work in line with best practice in the service. There are a range of national professional guidelines and patient pathways that represent standardized clinical practice in Norway, for example antibiotic guides, cancer care packages, and guidance plans for nursing practice. A prerequisite for safe use of language models in Norway is precisely that it is trained on data containing such national guidelines.
No systematic analysis has been done of the extent to which available language models are based on such knowledge.
Without specific training or guidance, language models can fall back on general or international knowledge rather than following country-specific guidelines and practices.
Practices and guidelines can be clarified and defined through standards, and they can help language models generate content that is in line with Norwegian conditions. An international survey may suggest that integration of standards in language models can improve their compliance with such standards [106].
Health administrative knowledge and practice in Norway
The Norwegian health and care sector is structured differently than in other countries. This applies to, among other things, the distribution between private and public health and care services and the geographical and administrative orientation. Furthermore, it also applies to arrangements such as general practitioners and the distinction between primary and specialist health services in general, as well as the central health administration [107].
A language model adapted to Norwegian health administrative knowledge and practice should be able to handle knowledge about this. It is currently uncertain to what extent available language models are capable of this [108].
It is important to recognize that health administrative knowledge and practice evolves differently than health professional knowledge, changing primarily through political decisions and legal modifications.
Health legislation in Norway
Rights and obligations of patients, relatives, healthcare personnel, and businesses vary from country to country. A language model adapted to Norwegian conditions also entails adaptation to Norwegian legislation.
A range of laws and regulations regulate the health and care sector, from general laws like the Health and Care Services Act to more specific laws like the Abortion Act. It is absolutely crucial that a language model in use in Norway does not provide information that violates either Norwegian law or Norwegian health law. No systematic analysis has been done of whether available language models are based on such knowledge.
Values and ethics in Norwegian health and care services
No language model is completely neutral when it comes to cultural preferences, ideology, and values. Research indicates that large language models reflect cultural values from training data. What utterances are perceived as offensive or aggressive will vary from culture to culture and will be based on the datasets used for training [109]. New large language models from other cultural areas than Western culture have also led to greater attention related to censorship and sensitivity and sparked a public debate. Commonly known, but politically charged, political events in China appear to have been censored in DeepSeek's language model [110]. Views on and knowledge about gender will, for example, vary between different countries and cultures.
The same will also apply to the health and care sector. In White Paper No. 26 (1999–2000) On Values for the Norwegian Health Service, it is stated that inviolable human dignity is the fundamental value for the service. Furthermore, equality, justice, equal access to services of good quality, professional responsibility, human dignity, and solidarity with the most vulnerable are central values.
This is also stated in Health Services in Norway that "[i]t is a fundamental principle that everyone in Norway shall have the same right to health services, regardless of age, gender, place of residence, and economy." [111].
In several white papers in recent years, emphasis has been placed on user participation as a fundamental prerequisite for good health and good care [112]. Patients, users, and relatives shall be seen and heard. User participation is a fundamental prerequisite in the patient's health service, where no decisions shall be made about you without you. It shall be based on the person's holistic needs, including physical, mental, social, spiritual, and existential needs.
Such values are not necessarily universal, and it must be ensured that language models in health and care services are in line with this value foundation.
[101] The new regulation implementing the language rules in the Sami Act will expand the range of nationwide organizations that Sami language users have the right to receive written responses from.
[103] https://www.regjeringen.no/no/tema/urfolk-og-minoriteter/nasjonale-minoriteter/midtspalte/minoritetssprakpakta/id86936/#:~:text=Nordsamisk%2C%20s%C3%B8rsamisk%2C%20lulesamisk%2C%20kvensk,mars%201998.
[107] i.e https://www.akademika.no/medisin-helse-og-psykologi/medisin-og-medisinske-disipliner/helsetjenesten-i-norge/9788205599253?srsltid=AfmBOoo6_oy48O6682W7srQUi7zMl0zIdtDXvyI1EMy3i94L8zT-pT6j
[108] ChatGPT 4o was queried in January 2025 about hospitals in Møre and Romsdal. When asked in English, it listed all four hospitals, but when asked in Norwegian, it only listed three hospitals, Kristiansund Hospital was omitted.
[109] https://arxiv.org/html/2402.10946v1 and https://academic.oup.com/pnasnexus/article/3/9/pgae346/7756548
[110] https://www.kode24.no/artikkel/4-grunner-til-a-vaere-skeptisk-til-deepseek/82594605 , https://www.khrono.no/fastlaste-verdier-for-alltid/939158
[111] Nylenna, Magne 2024 p.33