NLDB’26: A focused conference on NLP and information access offered insights and new trends

A small, but truly enthusiastic group of researchers and industry practitioners interested in applying Natural Language Processing (NLP) came together at the three day NLDB 2026 Conference in Trondheim in Mid-June. 

-    Two takeaways to me were insights on the dominant use of LLMs in computer science research. A plethora of models are in use everywhere within the domain. LLMs have rendered traditional methods mostly irrelevant. The panel discussed how increasing token costs highlight our dependence on Big Tech, says Associate Professor Benjamin Kille of host NorwAI in his initial summary of the conference. 

As local chair of the local organization, Benjamin Kille, was happy to welcome about  35 attendees to an intimate scientific gathering that invited participants to really make use of the networking opportunities by sharing projects, insights, and results. 

The first Annual International NLDB Conference took place in Versailles, France in 1995. Since its inaugural event, the NLDB conference has grown into a major specialized forum focusing on the application of Natural Language Processing (NLP) to database and information system. The annual event has been mostly in Europe, but took place in Japan last year, before entering the Nordic venue this year. 

22 papers accepted

The proceedings featured 22 peer-reviewed full papers out of 46 submissions. The exact titles of all 22 specific peer-reviewed papers are distributed directly within the newly released Springer LNCS Conference Proceedings. Professor Elena Cabrio at Université Côtes d’Azur and Professor Eric Monteiro at the host Norwegian University of Science and Technology edited the proceedings with three major themes: 

1. Long-Document Summarization & Text Mining

• The Problem: Standard LLM context windows frequently fail or hallucinate when processing multi-page industrial documents, legal texts, or massive datasets.
• Key Focus Area: Presentations focused on extreme compression ratios, introducing architectures that downsample long texts without losing critical semantic linkages or factual continuity.

2. AI Safety & Ethics

• The Problem: General-purpose AI models risk perpetuating data biases, violating user privacy during text mining, and consuming unsustainable amounts of power. 
• Key Focus Area: Researchers presented automated bias detection frameworks and privacy-preserving data extraction methods to evaluate frontier model safety metrics.

3. Sentiment Analysis & Web Data

• The Problem: Online user-generated data is heavily conversational, multi-layered (irony, sarcasm), and frequently polluted by automated systems.
• Key Focus Area: The conference showcased advanced algorithms tracking fine-grained emotional shifts in web text alongside modern techniques for filtering advanced web spam. 

Best paper award

-    The paper «Tokenizations for Austronesian Language Models: study on languages in Indonesia Archipelago» by Andhika Bernad Lumbantobing, Hokky Situngkir and Kevin Siringo Ringo won the award and a prize of 1000 Euros sponsored by Springer Publishing, says Benjamin Kille. 

The family of Austroesian languages are syllable-based as compared to most European languages whose tokenizers operate on letters and subwords. The author Kevin Siringo Ringo was present in Trondheim and received the price on behalf of the research team. 

The invited key note speakers were Erlend Aune from Cognite, the AI for Industry company, and Magnus Jensen of Schibsted AS, the Nordic Media company; both partners of SFI NorwAI.  

Published: 2026-06-25