Language and Personalization

LAP

Language and Personalization

This work package is a new construct, made up by the combination of the previous work packages on personalization (PERS) and natural language processing (LANG). The work in the work package will initially follow these two directions somewhat separately, for later to be tighter combined.

The purpose for this work package is to develop personalization techniques and Scandinavian language processing capabilities to provide personalized content generation and:

  1. Develop truly explainable, fair and transparent personalization techniques
  2. Enable proactivity in customer relations
  3. Provide an individualized experience that provably respects privacy concerns
  4. Develop individualized content
  5. Develop large-scale Scandinavian language models
  6. Enable human-like content creation and conversations

Personalization and contextualization have been successfully employed in diverse applications over the past decade, and currently see an extended usage, for instance in proactive interaction with customers and individualization of news stories. LAP will contribute to developing such systems while ensuring that the system usage will be ethical and respecting users’ requirements for privacy, fairness and accountability. 

Building Scandinavian language models requires the compilation of large-scale reusable language resources, including general-purpose corpora from public sources (e.g., news and social media) as well as industry- and domain-specific text collections. We will address the scarcity of the latter by pre-training on the former and developing transfer learning methods. These large-scale language models will then be utilized in real-life scenarios by formulating a number of specific summarization, explanation, and conversational tasks based on our partners’ use-cases. LAP will develop appropriate evaluation methodology with user-oriented evaluation measures and objectives. It will thus contribute to providing measurable quantification of the amount of domain-specific training material needed in order to provide a language service that is of sufficiently high quality.


Projects

Projects

LAP projects

Associated project

Associated project

TrustLLM - Democratize Trustworthy and Efficient Large Language Model Technology for Europe.

TrustLLM brings together leading European research institutions to lead European development in NLP and AI, and to lay the foundation for a broader European collaboration effort on LLMs and large-scale AI. The project envisions to build a series of LLMs to represent specific language families, in particular the Germanic language family.

Stories

Stories

-Vi treng generative norske språkmodeller

-Vi treng generative norske språkmodeller 

- Treng me generative språkmodellar på norsk? Svaret er eit rungande ja! Modellane må kunna formidla norske verdiar og haldningar, og dei må formidla god norsk - både bokmål og nynorsk.

Det sa Språkrådets direktør Åse Wetås da hun innledde på Trondheim Tech Port og NorwAI’s innovasjonsfrukost om språkmodeller og innovasjon 14. februar -24.

Åse Wetås on stage during a talk
Åse Wetås, Språkrådet. Foto: Kai T. Dragland

2024-02-27 


New language model for public use this winter

New language model for public use this winter 

-    NorwAI will meet the immense interest to work with language models with an open model, smaller thant that of the research model NorGPT-23 but will still be fully operational, says Jon Atle Gulla, professor and director at NorwAI.

Portrait Jon Atle Gulla
Jon Atle Gulla, Director of NorwAI

2023-10-06 


Upgrading infrastucture is critical to meet AI demands

Upgrading infrastucture is critical to meet AI demands 

Upgrading the national infrastructures are critical steppingstones to be prepared for the quantum leap technology now is facing.

Researcher by the Idun cluster
The Idun cluster is upgraded to meet the new demands of AI research. Here, NorwAI researcher Lemei Zhang.
Photo: Kai TY. Dragland, NTNU

2023-08-11 


Amerikanske språkmodeller påvirker ChatGPT. Det er problematisk.

Amerikanske språkmodeller påvirker ChatGPT. Det er problematisk.

-Flere problemer melder seg ved kunstig intelligens. Nå trenger vi å ta kontroll over infrastrukturen, sier Sven Størmer Thaulow, EVP og Chief Data and Technology Officer i Schibsted ASA.

Portrettbilde Sven Størmer Thaulow
-Flere problemer melder seg ved kunstig intelligens. Nå trenger vi å ta kontroll over infrastrukturen, sier Sven Størmer Thaulow, EVP og Chief Data and Technology Officer i Schibsted ASA.

2023-07-05


Schibsted reports on their AI results

Schibsted reports on their AI results 

Sven Størmer Thaulow on stage
Schibsted EVP, Chief Technology and Data Officer, Sven Størmer Thaulow was invited to the main media executive meeting in Norway to report on their initiatives in AI.

2023-05-15 


A national call for cooperation - Media to contribute to NorwAI’s LLM

A national call for cooperation Media to contribute  to NorwAI’s LLM 

Sven Størmer Thaulow on stage
Sven Størmer Thaulow reached out to his fellow media colleagues when he addressed Norwegian media executives at “Medialeder” – the yearly gathering of top media management in Norway on May 10th in Bergen.

2023-05-15 


Large language models as public goods

Large language models as public goods

Large language models have huge potential for value creation - but there is a strong need to address issues of control and risk mitigation.  

Portrait Eiviind THrondsen
Eivind Throndsen 
Academic coordinator  
Schibsted Products & Technology  

We are now moving towards a huge change in intellectual value creation, powered by the weird and surprisingly sophisticated mimicry of intelligence powered by large language models (LLMs). 

These models have unleashed a wave of creativity. They, and their model cousins that can process, transform and generate sound, images and any digitizable data,  have enabled previously impossible products and services along with a torrent of hype. 

Due to the enormous amounts of data, compute and brain power required, these important platforms are now mostly developed and controlled by a few very large private technology companies in the US. This is problematic, because along with all the interesting new functionality, large language models also suffer from serious and complicated challenges such as bias, hallucinations and toxicity. Private companies will invariably balance mitigating these issues with the need for profit. They are likely to do the bare minimum required to avoid regulatory retribution and public relations backlash.  

2023-03-28


ChatGPT and its inner workings

ChatGPT and its inner workings

Media insiders from seven countries got a lecture from NorwAI researcher Benjamin Kille as the interest to know more about the new language models dominated the discussions at a media lab day in Hamburg, Germany.

Portrait Benjamin Kille
Benjamin Kille, Post doctoral fellow, Department of Computer Science, NTNU
​​​

2023-01-31


Norwegian GPT model to be introduced

NorwAI to introduce large Norwegian GPT model 

NorwAI GPT Language Modeling Project is currently building its version of a large Norwegian model. The model will go into training this spring and will be ready for demonstrations for  interested partners,  says NorwAI head professor Jon Atle Gulla.

Jon Atle Gulla speaking on stage
- This will be a major result for NorwAI so far, says Professor Jon Atle Gulla. 
Photo: Kai T. Dragland, NTNU

2023-01-31 


The Kahoot Test of the AI Summary

The Kahoot test of the AI summary

Participants at the NxtMedia Conference 2022 were able to test journalistically written articles against summaries written by a language robot.

-The Kahoot game came in handy to choose the winners in the three examples, says adjunct associate professor Jon Espen Ingvaldsen who did the test.

Jon Espen Ingvaldsen holding a presentation
Photo: Kai T. Dragland, NTNU

2022-12-13


A new team of research assistants has started at NorwAI

A new team of research assistants has started at NorwAI

A new team of research assistants will continue our work with Kaia-The Social Robot

NorwAI will continue the research on Social Robotics. This semester three new research assistants have joined us and will develop new features and conduct extensive benchmarks to test Kaia against the state of the art.

Group picture of the research assistants
The team consists of Håkon Høgset (left), Alexander Gerlach (center), and Marte Eggen (right). PostDoc Benjamin Kille and Professor Jon Atle Gulla will guide their work.

2022-09-02 


Language experts to report on speech and text to the Storting

Language experts to report on speech and text to the Storting

Portrait Jonas Engestøl Wettre in the Technology Council
Jonas Engestøl Wettre,
project manager
in the ​​​​Norwegian Board
of Technology

The Norwegian Board of Technology (Teknologirådet) councels lawmakers and government. By starting with speech and later continuing on large language models, expert groups will disseminate the complex language technology step by step. NorwAI's director Jon Atle Gulla is part of the expert group.

2022-05-31

 


ECIR Conference

Norway may take a world-leading AI role 

Kjetil Nørvåg and Krisztian Balog at ECIR 2022 Conference
GENERAL CHAIRS –
professors Kjetil Nørvaag, NTNU (left)
and Krisztian Balog, UiS, headed the ECIR forum
for Information Retrieval in April.  Photo: NorwAI

-One particular area where the Norwegian AI stands out is the genuine interest in fairness, transparency and explainability, which align with societal values in Norway. Therefore, I can see Norwegian AI research taking a world-leading role in these areas, says professor Krisztian Balog at the University of Stavanger and Staff Research Scientist at Google. 

Krisztian Balog heads NorwAIs work package for language technologies. He cooperates with NorwAI's research director, professor Kjetil Nørvaag. The two professors joined their skills as general chairs of the successful ECIR conference in Stavanger during the week before Easter, giving an international audience insight on new research results in the broadly conceived area of Information Retrieval. 

2022-04-28


A silent challenge

A silent challenge

The Language Council of Norway has contacted NorwAI about current research on sign languages. There is ongoing research in Europe on AI-driven sign language processing, and NorwAI is considering looking into the use of machine learning for interpreting the Norwegian sign language. The visual and silent language is an official minority language in NorwAI.   Research will face some very special challenges if a project materializes.

Logo Språkrådet

2022-03-30


Tailoring news content: How Scandinavian mediahouses have tested recommender systems

Tailoring news content: How Scandinavian mediahouses have tested recommender systems

Portrait Jon Atle Gulla
Center Director Jon Atle Gulla
Photo: Kai T Dragland, NTNU

Scandinavian newspapers were early adapters to online services 25 years ago. Gradually some of them explored how recommender systems would enable individually tailored news streams. In an article in AI Magazine recently NorwAI associates, headed by Center director Jon Atle Gulla (picture) explore how Scandinavian media organizations are coping with these new technological opportunities.

2021-12-20


New Language Models in NorwAI

New Language Models in NorwAI

Photo. Jon Atle Gulla

The NorwAI center is determined to provide new Norwegian language models that are significantly larger and better than what is available to-day and can easily be employed in advanced Norwegian NLP applications for industrial use, says center director professor Jon Atle Gulla.

2021-04-20