Remi van Trijp

Back in my early twenties at the University of Antwerp, one of my professors told me: "Your stylistics is better than your linguistics." He meant it as a compliment — he really liked one of my essays — but at the same time it felt like a kind warning to major in literature instead. In hindsight, it's probably the most accurate review I've ever received, because I was indeed more interested in how language evokes meaning and emotion than in dissecting it into tree diagrams.

And yet, a year later I found myself disregarding his advice. I loved literature, but I had always wanted to do computer science as well. One seduced me with metaphors and narratives, the other with systems, structure and the thrill of building things.

Computational linguistics seemed the closest thing to both — so I signed up for the linguistics programme.

I remember stepping into the office of Walter Daelemans, a pioneer in Natural Language Processing, on the very same day that Lernout & Hauspie — Belgium's greatest language-tech hope — was declared bankrupt. Walter told me that my chances of building a career in language technology, at least in Belgium, had just greatly dimmed. But I was too young and naive to have thought about the money — let alone a career.

In Walter's introductory class to Artificial Intelligence, I first encountered the Talking Heads experiment of Luc Steels — and it instantly changed my life. Everything clicked. Luc's robotic agents weren't just parsing text: they were truly communicating about the world. They invented words, negotiated meaning, failed gracefully at first but inevitably self-organized their community language. The Talking Heads Experiment wasn't just clever. It felt alive.

That moment opened a new vision of language to me. Not as a formal system of rules, but as a living ecology of form and meaning. And I wanted to understand how that ecology works.

From Friction to Flow (2005—2020)

Then I got lucky. Very, very lucky. After graduation, and with the local language-tech bubble burst, I worked for some time as a journalist and as a language teacher in secondary school. But then I learned from Walter that Luc Steels was looking for new people. He had already hired roboticists and computer scientists, and now he was scouting for a linguist who wasn't afraid of computers. And those were still in short supply back then. Luc invited me to his lab in Brussels, and after what must have been the worst job interview ever (I really thought I was there just to visit), I landed a position in Paris where he directed the European branch of the Sony Computer Science Laboratories.

Luc hired me on a three-year contract for a European project. More than twenty years later, I'm still there. At some point, I stopped being a guest and started being part of the furniture.

Like many linguists in a computational world, I felt completely useless at first. My colleagues were building sophisticated systems and running simulations. I could barely get a sentence parsed. It took me six months to produce anything that didn't break. But I dug in. Learned that as a linguist, I had a different perspective. And over time, I found what I could contribute.

I helped develop Fluid Construction Grammar (FCG) and the multi-agent framework Babel it is embedded in. More a musician than an instrument maker, I kept pushing at the edges of our formalisms to see which sounds I could coax from them. If you came across FCG in the mid-2010s and found it quirky, I probably had something to do with it. My colleagues later did a tremendous job at professionalizing the software — most of the stranger decisions got refactored into more elegant code. But if you look deep enough under the hood, you can still find some of my fingerprints.

I had my reasons. Probably.

I also became a Master of Puppets. For my PhD, I conducted multi-agent experiments about the emergence of a grammar for describing argument structure — who does what to whom. My agents would observe actions in a puppet theatre and develop a language for talking about those events.

To understand how such grammars function, I began using FCG for reverse-engineering how real languages solve the same problem. It was an exhilarating time: no one had actually implemented a Construction Grammar approach to natural language that worked for both production and comprehension. The community was brimming with innovative ideas, but without a working model, Construction Grammar was often dismissed as linguistics on holiday.

I'd like to think I helped prove that feeling wrong.

The Narrative Turn (2020—now)

In my recent work, I became interested in what is perhaps our most powerful cognitive tool for making sense of the world and each other: narrative.

Narratives shape how we interpret, organize and assign meaning to events. We weave stories instinctively, turning fragmented experiences into coherent structures, causal explanations and temporal frames.

This shift of perspective has, again, profoundly changed how I think about language, Artificial Intelligence, and even about the nature of science itself.

Each sentence, for instance, is a mini-narrative in its own right — mirroring the larger discourse structures in which it is embedded. And every language user has this talent of creating cinematic experiences. If I say the cat chased the mouse, I am directing a fast-paced action flick. But if I say the mouse was chased by the cat, we flip the camera to the mouse, and suddenly we are in a horror film — watching a poor rodent try to escape from a looming threat.

In both cases, the same story has inspired two entirely different blockbusters — each shaped by the spin the Active and Passive constructions put on the event. How we go from the story (a conceptual scene) to a linguistic movie's final edit (the sentence) — and why we choose one framing over the other — is not just a matter of syntax. It's a lens into cognition, perspective-taking, and communicative intent.

This matters for not just for the study of language, but for Artificial Intelligence and technology as well. Meaning isn't just what a sentence says; it's what it does. The same event can be framed as action, victimhood, achievement, or tragedy. And humans are exquisitely sensitive to these framings: we infer intentions, assign blame, shift sympathies, or reshape decisions based on subtle grammatical cues. An AI system that overlooks this won't just sound robotic: it will never understand how people construct shared realities.

I believe this is why the scientific study of language is more relevant than ever — even in a world where large language models seem to become evermore powerful. If we want AI to truly center on improving people's daily lives and benefit society, it has to move beyond data-driven generation and into intention-driven production and comprehension. It needs to recognize not just what was said, but also how it was framed: what was left out, what was foregrounded, and what kind of story the language user is inviting you to inhabit. This is more than a linguistic or technological challenge. It is a socio-cognitive one. And a deeply human one.

Narratives are also indispensible for the scientific enterprise. In recent years, I've become aware of a tension that many researchers now feel. On one side, the traditional image of science as the quest for objective truth: systematic, generalizable, detached. On the other, deconstructivism, which dismantles foundational assumptions, questions categories, and exposes the hidden politics of objectivity itself. These two visions seem to pull in opposite directions: one building towards control, the other unraveling towards epistemic humility.

Narrative-based explanations may offer a way to reconcile them. Science is often tasked with explaining non-recurrent phenomena on the basis of evidence that is fragmentary, indirect, and incomplete. In such contexts, narratives aren't optional; they're integral to how plausible, coherent and causally structured explanations are constructed. Longtime dismissed as "just-so stories" — intuitive, post-hoc reconstructions rather than genuine vehicles of knowledge — narratives are increasingly recognized for their role in epistemic knowledge-making.

In my work, I try not only to study narrative; but to embody it. My writing style has become increasingly narrative-shaped — a modest attempt of trying to map scientific thinking, and to explore how that thinking unfolds in the same way that all meaning does: through interaction, the meeting of minds, and situated context.

I don't expect my experiment to persuade everyone, but I do hope it's not lukewarm. May it leave you hot or cold, intrigued or infuriated — but never indifferent.

This is where I am right now. But ask me again in five years. I'll have a different story to tell.