14/05/2026
It was a real pleasure to take part as an invited speaker at the 10th Workshop on Linked Data in Linguistics, co-located with in Palma de Mallorca. My talk, titled Making Knowledge Visible Again: Linguistic Linked Data in the Age of Large Language Models, explored a key question: What happens to knowledge when language technologies are probabilistic, opaque, and parametric?
Over the past decade, NLP has undergone a profound paradigm shift:
from explicit linguistic knowledge — lexica, ontologies, knowledge graphs — to massive probabilistic models trained on web-scale data.
LLMs are extraordinarily powerful.
But they also raise critical questions about:
🔹 knowledge grounding
🔹 traceability
🔹 interpretability
🔹 epistemic control
One of the most critical phenomena discussed was the so-called model collapse, when AI systems increasingly learn from AI-generated content, progressively flattening linguistic diversity and marginalising rare forms, minority languages, and cultural nuance, as discussed by my colleague Eva Vanmassenhove .
This is why Linguistic Linked Open Data (LLOD) matters today more than ever, in my opinion.
Not as an alternative to LLMs — but as a complementary semantic infrastructure capable of providing:
✅ explicit knowledge
✅ provenance and traceability
✅ multilingual grounding
✅ semantic disambiguation
✅ interpretable AI workflows
The future of trustworthy AI is hybrid, as recent studies have discussed, we need to move:
➡️ from black-box scale to semantic grounding
➡️ from web scraping to curated knowledge infrastructures
➡️ from probabilistic fluency to interpretable meaning
Because making knowledge visible again is not only a technical challenge but also a cultural and epistemological responsibility.
