Roberto Zamparelli

After a PhD in Rochester, N.Y. and a Marie Curie postdoc in Edinburgh, Roberto Zamparelli has worked in Italy, first at the University of Bergamo, then in Trento, where he was a founding member of CIMEC (Interdepartmental Center for Mind and Brain Sciences, University of Trento). Zamparelli started his career as a theoretical linguist working on the syntax/semantics interface of nominal structures, developing a theory that combined DP cartography with compositional semantics ("Layers in the DP", 2000). With prof. C. Heycock, he looked at the syntax and semantics of (bare) noun coordinations; in other collaborations, he wrote papers on the count-mass distinction, NP pro-forms and partitivity. Since 2010 he has been interested in computational linguistics and its connections with formal semantics ("Frege in Space" 2014, with Marco Baroni and Raffaella Bernardi) and on the use of artificial neural networks to probe the innatism/empiricism debate (joint work with Cristiano Chesi, Shammur Chowdhury and others). More recently, he has worked on games for teaching linguistics structures to a broader audience, or to collect and validate data on the prosody-semantics mapping. His current project uses ERP to analyze the difference between listening to sentences and mentally composing them.

‍

Interview

OASIS: Thanks for joining us for an interview. What's your favorite ontological entity and why?

RZ: I have a weak spot for subkinds...😍 My advisor was Greg Carlson, who wrote one of the PhD theses with the longest lasting influence in semantics. And I have always been fascinated by the different ways languages use to refer to “kinds of things”. Why SUBkinds, then? Because we seem to be very good at switching from token-talk to kind-talk (think of what we are really saying in “Every car sold in the US undergoes thorough crash tests”) and because we can break kinds down into subkinds in many ways, which cannot really be counted together (“??There are two kinds of dog in that room: a guard dog and a German Shepard” — an old example from Carlson). I am also intrigued by their anaphoric properties: after a previous mention we can pick up tokens with “the” (“a dog came in… the dog…”) but it takes a demonstrative to pick up a kind (“Dogs are common … this type of pets …”). Why? As far as I know, nobody has a comprehensive theory.  

OASIS: What is everybody wrong about?

RZ: We spend a lot of time folding structures in intricate ways and not enough smoothing the folds into workable trees -- in a principled way! I have been working in correlative constructions lately, and there is a lot of evidence that they involve elaborate movements of structures and remnants, which create left-expanding structures. Put a few of these together, as in “Both Sinner and Alcaraz have both challenged and defeated both Djokovic and Federer” and you get a structure which makes Baroque churches look like Ikea tables. There must be a way where structures which are the result of complex movements are simplified (one word could be “reanalyzed”, but it has a whiff of performance deficits that I don’t like), and nobody talks about this!

OASIS: Who are your professional heroes?

RZ: Carlson, Kayne, Steedman. Boring, you could say! But they are truly great linguists, by which I mean people capable of seeing data under everybody eyes in a completely different light, of convincing you (at least for a while..) that everything you thought of as established is wrong; of showing that there is a way to reconcile the computational implementation of a linguistic theory with the idea that it should provide real insight on what is possible in the human mind. We need more people like them, but they don’t come frequently.

OASIS: What would you like to tell students/early career researchers?

RZ: Don't just use ChatGPT, keep both eyes open on what people (painstakingly) discover about how it works. Theoretical linguists cannot afford to ignore computational ones, and they_CL need us_TL even when they think they don't. There is a lot of ground to cover before CL starts to approximate a model of the human language faculty – the size and manner of training is too diverse, the output too inconsistent, superhuman in some corners, baboon-level in others. But as methodologies start to align, we could definitely use computational models. Also, mapping network states with brain states is a fast growing subfield, and it could become very interesting.   

OASIS: Tell us how your career trajectory turned out in an unexpected way.

RZ: I wallow in detours. I started university studying modern Italian literature, ended up getting a degree in psycholinguistics after a stint with formal logic. In the US I studied with Tom Bever, a psycholinguist, but graduated with Carlson. Later, I have mostly worked on the syntax-semantics interface, but my most cited article is one with Marco Baroni in computational linguistics. I have created physical educational games of syntax and online games to collect prosodic data. All this not because I am particularly smart, only because I am incredibly curious and equally incredibly easily distracted. What was your question, by the way?

OASIS: Do you work step by step or follow hunches (or neither, or both)? Explain?

RZ: Hunches, often under the shower, then I search for the flow. The shower is the easy part. 

OASIS: Tell us about a student you've learned something from.

RZ: Doing cognitive science, it happens weekly. Lastly, Dota Dong and Giulia Li Calzi --- little suns made into people. Often, it involves computational techniques they have picked up elsewhere, or ideas on where to apply them. I have been lucky enough to be on the thesis committee of many interesting people, and I learned a lot from them.

OASIS: Your PhD thesis in six words.

RZ: Noun phrase cartography meets compositional semantics.

OASIS: Thanks for talking with us. Looking forward to seeing you in Edinburgh! 

RZ: Looking forward to being there!

‍

‍