Annual Review of Cognitive Linguistics 6 (2008), 329–344. doi 10.1075/arcl.6.18ste issn 1572–0268 / e-issn 1572–0276 © John Benjamins Publishing Company Interviews A whole-systems approach to language An interview with Luc Steels Benjamin K. Bergen Department of Linguistics, University of Hawaii, Manoa BB: You are best known for your work in Artiicial Intelligence and Robotics, where you have been instrumental in creating entirely new paradigms, like Behav- ior-Based Robotics and Semiotic Dynamics. However, many cognitive linguists have only recently come to know about your work through your development of Fluid Construction Grammar (which we’ll talk about later in this interview). As a consequence, I think most readers will be surprised to know that your early training was actually in linguistics. Could you tell us about your early linguistics training, how it led you into Artiicial Intelligence, as you conceive it, and how you found your way back to linguistics? I wonder how your current view of language has been colored by this journey. LS: Yes, indeed, my irst love as a student was for linguistics. It was the early seven- ties and the Chomskyan wave was sweeping over Europe. I was a student at the Uni- versity of Antwerpen (in Belgium). We were the irst generation at a newly founded university and so our professors were young and highly motivated. hey brought us in contact with all the new trends in the ield, principally generative grammar and Montague semantics. Fortunately, they also taught us about the European tradition of language studies, which has always been more oriented towards a functional, meaning-centered view of language as opposed to a syntax-centered structuralist view. We also had courses in historical linguistics in the tradition of the philologists, courses in the newly emerging ield of sociolinguistics, the philosophy of language, phonetics, etc., giving me a very broad solid basis in the whole ield of linguistics. In my last undergraduate year, I discovered on my own the power of computers for operationalising and testing linguistic theories. his was not so obvious because computers were still very rare. here was only one central computer for the whole university and it was being programmed with punch cards that you could submit twice a day. Doing language processing with these brute machines was unheard of. Whereas most linguists at that time and still today focus on describing lan- guages as they are, I became interested in the question of what kind of information