Stephen Wilson 0:06 Welcome to Episode 24 of the Language Neuroscience Podcast. I'm Stephen Wilson. Thanks for listening. My guest today is Liina Pylkkanen, Professor of Linguistics and Psychology at NYU. Liina investigates the neural basis of semantic and syntactic combinatorial processing using magnetoencephalography, or MEG. Much of her work is characterized by elegant, carefully constructed experimental designs that are informed by linguistic theory. And she really takes advantage of the temporal and spatial resolution of MEG, to study neural responses to specific linguistic operations. As an example of this kind of work, we're going to talk about her recent paper, ‘Disentangling semantic composition and semantic association in the left temporal lobe’, with first author Jixing Li, Journal of Neuroscience, 2021. Okay, let's get to it. Hi Liina. How are you? Liina Pylkkänen 0:55 Hi, Stephen. I'm pretty good. How are you? Stephen Wilson 0:58 Yeah. Good. It’s a lovely day in Nashville. How about you? Where are you at? Liina Pylkkänen 1:03 I'm in lower Manhattan at home and it is basically a summer day here too, which is not usual for early November. But yeah, that’s where we are. Stephen Wilson 1:13 I'm really hoping that this warm spell will just last a little bit longer. I want to, I want to like kind of fit in one more camping trip before the winter is upon us. (Laughter) Liina Pylkkänen 1:23 Yeah, it's definitely going to be a very nice weekend here. Stephen Wilson 1:26 Yeah, so we've been trying to set this up for a long time. I think there was like, multiple bouts of COVID and then summer plans and all kinds of stuff, right? Liina Pylkkänen 1:35 Yeah, it's been in the works for a while. But finally, we have made it. Nobody has COVID as far as we can tell. (Laughter) Stephen Wilson 1:41 So, I know that you've listened to the podcast before. So, you know that I’d like to get started by asking people about their interests when they were a kid, and how they found their way into this niche of science that we all find ourselves in. So, can you tell me about that for you? Like, what were you interested in when you were a kid? Liina Pylkkänen 1:41 Yeah. So I definitely knew that I wanted to somehow make language my job very, very early on. So, it kind of clicked for me, actually, when I first studied, started studying English in school. So, in the Finnish system, back then, it was at the start of third grade, so I was eight years old then. Stephen Wilson 2:26 Were you speaking only, only Finnish up to that point? Liina Pylkkänen 2:29 That's right. I was only speaking Finnish up to that point. And I remember how I just immediately kind of love the feeling of a foreign language in my mouth, sort of. And I also remember that as we started learning vocabulary, like we would just point to objects in the room and start learning the words for those objects, I felt that I needed to decide whether I should have an American accent, which is what I mostly heard on TV, or the British accent that my teacher had. And no other student was like wrestling with this question, but I felt like well, now I need to like pick the phonology. And since I'm, mostly heard American English on TV, I went with that. And the teacher was kind of puzzled, is like, well, how do I all of a sudden have become an American girl? (Laughter) But then, that was already sort of an early interest and just kind of learning a foreign language. So that's sort of what it was like, up until I got to university. So I just really enjoyed, you know, learning foreign languages, I did the kind of usual European menus. I had four foreign languages in school, there was nothing exceptional about that. And that, what, what exactly you do with that interest is a little unclear. So I definitely wanted language to be the job, but how to do that, you don't, you don't know. Right? Very canonical choice in many European universities, is to then major in some type of philology in university. So then I was admitted into the English philology major after I graduated high school, and there, it's a combination of literature, and then you do a little bit of linguistics. And I immediately knew that it really, I really had no particular interest in doing literature. And as soon as I encountered linguistics, I immediately knew that that was, that is really what I was looking for. Stephen Wilson 4:29 That was it. Liina Pylkkänen 4:30 Yeah. And then, you know, I had kind of wanted to do an exchange here in the US already as a high school student, but that would have been extremely expensive. But then when you are in university, they have exchanged programs with US universities. And so then as soon as I was able to do that, I applied for one of those fellowships I did, and then I got one and they, I ended up in the University of Pittsburgh where there was a really good linguistics department at the time, and I was able to sort of get started. Stephen Wilson 5:01 Okay, what was Pittsburgh like as your sort of introduction to the English speaking world? Liina Pylkkänen 5:06 I really enjoyed it. So, it was a bigger city than the city that I was coming from in Finland, which was Tampere, which is the second largest city in Finland, but it's still not a large city. So, Pittsburgh was a more exciting place and I had had this kind of desire to be in the US for a long time and so then I really felt like I had arrived. And I just felt, you know, very comfortable right away. And, you know, just had a really great time. And the department was a really, really nice environment. And so I was an undergraduate exchange student at that point. But I was able to do a lot of the Masters classes already during that exchange year. And then, through a kind of a complicated path, I was able to get admitted to the master's program, even though I did not have an undergraduate degree yet. So then I did an MA there in Linguistics before I went to do my PhD at MIT. Stephen Wilson 6:05 Did you ever finish your undergrad degree? Liina Pylkkänen 6:07 I didn't finish any type of undergraduate degree. So like, if you look at my CV, it kind of looks like I'm a college dropout, but um, but also at the time, in the Finnish university system, there really wasn't anything like the US undergraduate degree, which was also common in other Euro, European systems. So you, Masters really was the first degree that you were going to do anyway. Stephen Wilson 6:30 Right. And did the, did the Americans accept your American accent? Did they find it authentic? (Laughter) Liina Pylkkänen 6:38 Yes. I actually have a funny story. The very first day that I had arrived at Pitt, University of Pittsburgh, I was in the elevator in the Cathedral of Learning, so there's the the main building on campus is a very tall building. I forget how many floors now, but very many, and linguistics was like up there. And I was in the elevator and I was chatting with somebody and somehow it came up that I was an exchange student and then that person said, like, oh, you really don't have an accent? Like, how long have you been here? Well, I came yesterday. (Laughter) That's crazy. I'm like, well, but I've sounded like this more or less since I was eight, when I started obsessing about pronunciation. Stephen Wilson 7:23 That's funny. Liina Pylkkänen 7:24 Yeah. Stephen Wilson 7:25 Okay, so you went to MIT, like the epicenter of linguistics, or at least one kind of linguistics, I guess I should say. And how was that? Did you have a good time there? Liina Pylkkänen 7:37 I really did have an amazing time at MIT. It was a very marked choice for me for graduate school, because the Pitt Linguistics Department wasn't or at least my advisor there who was rich, rich Thomason, kind of one of the Montego Bay in St. Francis's was not a super generative Chomsky in place. And so I actually thought and the kind of research that I was doing, really wasn't, so I was using a formalism that I've kind of stitched together from, you know, lf chi and hbsag, which is not, you know, the sort of Chomsky in theory. And so I really thought I was gonna go to like Stanford, or Ohio State or one of those universities. But then when I visited MIT, it was the place where I felt like I kind of least understood the way they were thinking. And, but it was clearly the kind of dominant paradigm and I, at that point, like, amazingly, for my, whatever, 23 year old brain, I, I have kind of an inside it was like, if I don't come here, if I don't study with these guys, I'll never really understand this. And then I also really love the student body, like they, you know, they were really amazing. So I chose so then I chose MIT. And I really, I mean, those years, I would still say are, you know, one of the best in my life. So it really was a special time. And I really, really did enjoy it. Stephen Wilson 9:10 Yeah, but you didn't really grow into like a hardcore generativist, did you? Liina Pylkkänen 9:14 Um, I mean, I, you know, I'm an empiricist. I, you know, I did a fully theoretical dissertation, I was always interested in the syntax semantics interface and I came into the program with a pretty good training in kind of event based semantics. And then at MIT, I encountered a, you know, kind of a more formal way to do compositional semantics. And then I sort of married the theory that I had learnt really during my Masters at Pitt, with the more Heim and Kratzer type of, you know, lambda calculus of compositional semantics and then kind of worked with that system, and Angelika Kratzer was, you know, had also been developing that, that combination. And so, so I worked, you know, within a lot of the typical assumptions, you know? Stephen Wilson 9:14 Yeah, so your dissertation is not about, it's not a cognitive neuroscience dissertation? Liina Pylkkänen 10:41 No. Stephen Wilson 10:41 I did not know that. Liina Pylkkänen 10:24 No, it's, my dissertation is called ‘Introducing arguments’ and it gives a theory about kind of the verbal domain, specifically how arguments of verbs other than, say, the subject, and the object or kind of the core arguments get introduced and to the meaning. So, for example, I worked a lot with indirect objects, also cross-linguistically in African languages and causative constructions like ‘bill broke the glass’ type construction. So that's still my most cited work, Stephen. (Laughter) Stephen Wilson 11:01 I didn’t know that. Yeah, I used to be very interested in argument structure too, when I was like a student. I was trained by people that were very into argument structure. So yeah, I'm sure we could probably talk about that for a whole hour if you want to do but I don't know that everybody would appreciate that. So, you started doing MEG or Magnetoencephalography, which seems compulsory because you're Finnish, right? And Meg is like the domain of Finns. Now, is there like a national rule that, that’s like the methodology that you have to use when you're Finnish? Liina Pylkkänen 11:36 No, that is the running joke that I am the control case. So even if you get displace from Finland, you still end up doing MEG, because it's just like programmed in you. But no, it is interesting. So I am Finnish and I do MEG but I really did not train with, you know, Finnish Pioneers. I really discovered MEG at MIT. It took me a while before I learned that the Finns do it too. Stephen Wilson 11:59 So, why, why it is? I mean, yeah, I was kind of joking. That is really funny. But why, why is like Finland, so strong in MEG research? Were they critical, sort of pioneers of the technology over there? Liina Pylkkänen 12:11 Yeah. So I think they were critical pioneers of the technology and then, you know, there's just certain pioneering researchers who, you know, really started showing the utility of the technique for, you know, kind of more cognitive questions, like Riitta Hari and then, in the more, you know, analysis techniques side, Matti Hämäläinen and so, and there's, you know, there's a couple of really thriving groups that do magnetoencephalography. So, yeah. Stephen Wilson 12:43 So can you tell me about how you, like when you first encountered that methodology, but it's very exciting to realize that you're going to be able to address your questions in the brain? Liina Pylkkänen 12:55 Yeah. So, that's not at all how it started. So I came to MIT with this really interesting formal semantics, because that is what I had, you know, have always been interested in. But then, as the first summer came, so after the first year on the program, I needed a summer job, there really wasn't summer funding, so I needed to do something and I didn't want to go back to Finland, and you know, do something totally unrelated. And at that point, Alec Marantz had acquired an MEG machine that, you know, he controlled, and it was pretty new. But since I was really new too, I didn't really understand how new it was. So you know, like, he was also like, from sort of, still very much learn about it. And he was looking for kind of a lab manager RA just for the summer, and I had no interest in experimental research at all. It sort of had like an unappealingly practical flavor for me, because I’d sort of thought of myself as very theoretical and more, you know, leaning towards philosophy than anything as practical as psycholinguistics or neurolinguistics, but I really needed a job. So then I was like, okay, I can try to like push these buttons. And so then I came into that position, knowing nothing, and just gradually kind of started to learn. And then something actually kind of clicked. So the next year, I did then, you know, take neurolinguistics and I started getting into it more intellectually. And as I think about it backwards, I think what really appealed to me, is the notion that you can have a result. So when you're doing theoretical research, you never know when a project is finished, like there just is no definition of when you are finished with a project. When you're doing experiments, there is a way to define a project and there is kind of a way to know when you're done. And that suits my personality way better than this feeling that you're never done. And so I think, you know, like I did, you know, I was very intrigued by the brain, like you feel like kind of an astronaut, you're going to unknown places, you know, if you do manipulation that, you know, nobody has ever done, it's incredibly exciting to see how that affects the brain because like, nobody has seen that before. It's like traveling to a new new city every time. I, you know, so that was important, but the notion of like being able to complete a project and kind of organize your life more in that way, it was definitely something that, that appealed to me. Stephen Wilson 13:35 Hah! And so that was when it sort of, your research started to go more and more in that direction. Liina Pylkkänen 15:39 So then, in graduate school, I then had these two lives. So I was doing theoretical work very seriously. But then I also started to work on just basic lexical access, not because I was interested in lexical access, but because at that point, it really felt that we kind of knew almost nothing. So I'm talking like ‘98, ‘99 now, so a while ago, um, and you couldn't ask questions about applicative and causative constructions or any of the types of questions that I really was interested in, theoretically. And it seemed like, if you want it to kind of carve out to the brain space, you have to have some clue of what just like basic lexical access would look like, in the brain, in order to start getting into the more structural aspects of it. So obviously, like, you could be interested, more interested at lower levels, like, you know, phonological categorization, or something like that. But I was always a meaning person, not a sound person. So it seemed like the, you know, like lexical access was sort of the starting point. That was types of questions. Stephen Wilson 16:52 And, obviously, you've kind of brought them together over the years, right? I mean, you've made your research program more and more about combinatorial semantics. Liina Pylkkänen 17:04 Yeah, so it took a while. But then when I, you know, started as a postdoc at NYU, that is, when I was, you know, when I was kind of, for the first time, able to connect with my kind of theoretical expertise. So, I had kind of a floating postdoc across labs, NYU was kind of working on a faculty position for me, and they first were able to bring me in as a, as a postdoc. So I was working with Brian Makary, who had done a lot of behavioral work on so called corrigin constructions. So expressions like the ‘author began the book’, where ‘begin’ and ‘book’ don't really easily go to gather semantically upon first sight, because ‘begin’ is kind of an aspectual verb and it’s really looking for something like an events predicate to combine with like ‘begin singing’ or something like that. But when we get something like ‘begin the book’, there's actually a decent amount of behavioral evidence that something extra happens there. There's like a processing delay and we think it's associated with some kind of an invoke, it, like you're invoking that implicit activity. So you are, in fact, interpreting that expression as the author began some activity involving the book. So we, you know, then did a first MEG experimental, those kinds of expressions. And then I did, you know, like a bunch of work on different types of expressions of that flavor. And that was kind of like the first body of work that was really that kind of on representational questions that I really was also theoretically interested in. So yeah, I would say that's the postdoc level. You know, it was sort of a little more unified, but I was not able to continue my theoretical work, I really wasn't able to, like do like, cognitive neuroscience of language seriously and then I also theoretical. Here is like, I mean, this… Stephen Wilson 17:11 No. This is a full time job. I mean, unless you like, you know…. Liina Pylkkänen 19:01 Yeah, yeah. Something would have to give. Yeah. Stephen Wilson 19:13 So, how would you I mean, like, for somebody that doesn't know, your, you know, what your lab is all about? Like, how would you sum up like the sort of central questions of your lab these days? Liina Pylkkänen 19:26 Well, I think I've had the same central question for a really long time. And it's a it's a really hard one. So I may always have it, I doubt that we're really completely going to crack it and mechanistically but it really is the question, how do our brains construct complex meanings? Of course, that is, in some sense, the question that everyone is asking. But, um, I think, you know, so, you know, ultimately the answer will be a description of you know, like, the full machine that's in the brain that accomplishes in it, and that accomplishes this, and hopefully we'll have, you know, a very mechanistic understanding of it. I think what we have been able to do, is, you know, shed some detail into the picture. So, obviously, a big focus of our studies for the last many years has been the function of the left anterior temporal cortex, which has been kind of a most consistent correlate of composition in our hands. And since we're studying it with Magneto encephalography, we're able to kind of lock the analysis into a certain timeframe. So we're able to study a certain stage of processing as opposed to just a spatial location. And I think that's very important, because that has allowed us to have at least you know, some level of consistency in the responses across study. So we know that we're always like looking at this relatively early response, around 200 to 250 milliseconds after the onset of usually a visual word that's in a combinatorial context. Basic finding is that there's an elevation and amplitude, you know, if the word is in a combinatorial contacts, and it does not seem to, although this was our first guess that it might reflect something like syntactic composition, I think we've pretty compelling, compellingly ruled out that hypothesis, and it does seem to have a more conceptual flavor. So it is as some variants of conceptual combination. And we've done a pretty systematic body of work, kind of really trying to understand what are the possible inputs to that, that operation and what are not possible inputs to that operation? So you know, it has sort of a level of detail that I find satisfying in the sense that we're not, it's not just like, oh, even not just this is conceptual combination, but we're able to say something about it in that in a little bit more detail. So it goes back to that kind of theoretical work where you're trying to actually explain, you know, why a sentence means what it means. So, you know, theoretical semantics or formal semantics we, we write the lexical entries of each of the words, and there's a formalism that puts those lexical entries together. And then there's a sentence meaning that is compositionally derived from the pieces. And so like, at the end of the day, we'd love to do something like that at, but the, you know, as a neural description. And we're very far from that, obviously, very far. But, you know, it's like, when we start to understand the details, that's what I find interesting and sort of fun. Yeah. Stephen Wilson 22:58 Right. Yeah. I mean, I think that your lab has like one of the most cohesive focused research programs of any lab in our field, you know, like, it's like, you know, you definitely have a central question. And everything that you do is revolves around that. And, and I think you take this very sort of bottom up approach to which I think is interesting. What I wanted to do is talk about one of the recent papers that we had mentioned, and that kind of give the give our listeners like a an example of this work. The one I thought would be a good fit is called ‘Disentangling semantic composition and semantic association in the left temporal lobe’ by Li and Pylkkänen. So can we like, kind of delve into that one a bit? And maybe first, could you tell me a bit first about your co author on this paper? Liina Pylkkänen 23:49 Yeah. So Jixing Li, was a postdoc in our Abu Dhabi branch of the lab. She did her PhD with John Hale, who was at the time was still at Cornell. And now actually, she just started a faculty position in Hong Kong. So she's just first semester assistant professor. Stephen Wilson 24:11 Yeah, so it's a very nice paper that that she's first author of in the Journal of Neuroscience. And before we like, kind of get into the details of it, you know, you talked about MEG (Meg), I guess you call it MEG (M.E.G.), I don't know. I call it MEG (Meg). Is that acceptable? Or is that like a faux pas? Liina Pylkkänen 24:27 I think we would have to say MEG (M.E.G.). I actually don't find MEG (Meg) very acceptable, but…. Stephen Wilson 24:31 Really! Okay, fine. I'll try. Okay. I’ll try. MEG (M.E.G). People do say MEG (Meg). People do say it? Liina Pylkkänen 24:37 They do say it, it's fine. It's fine. Stephen Wilson 24:39 You don’t like it? Okay, well, I'll see what I can do. So, you mentioned the kind of the, the ability to ask temporal questions. So can you kind of talk for people that are not familiar with the methodology? Can you talk about the spatial and temporal resolution of it and kind of how it works and the advantages in that respect? Liina Pylkkänen 24:59 Yeah, so MEG reflects the magnetic fields that are associated with, associated with our neurons electrical activity. And so it has millisecond temporal resolution. So as quick as you want to go, we can go with, with MEG. So it's parallel to EEG in that sense. But then unlike EEG, which kind of gets, you know, the potentials get distorted as you go from the neuronal source, by the time you're at the scalp, there's sort of, you know, it's not a pretty reflection of what was the underlying configuration. So the relationship between the magnetic fields that we capture outside the head, and the configuration of the sources in the brain is just much more transparent for for MEG than it is for EEG. So we're able to create a model of the sources. And so it is important to keep in mind that we're not imaging the brain, it's not an image of the neuronal activity, it is a model of it. And you know, as with any kind of modeling, like it'll change if you do it with a different pipeline or something like that. So, so it's rougher but the spatial resolution is in the order of kind of centimeters, depending on where you are inside the head. Deeper sources are less accurate than more superficial sources. That's one of the differences. But, but it's you know, as for how non invasive methods go, it is the best combination of time and space that we have non invasively. And so if you want to have a measurement of the whole head, with really great temporal resolution, and pretty good spatial resolution, MEG is the way to get it with what we have available. Stephen Wilson 26:51 Yeah, it I mean, it's the advantages are clear, and not many people, there’s not that many universities that have it, right? I mean, it's nowhere near as common as functional MRI or EEG. Liina Pylkkänen 27:03 Yeah, hasn't taken off clinically the way that you know, MRI has and so that's…. Stephen Wilson 27:10 I think it's just a lot more expensive, right? And, and, well, you don't think so then, I guess a scanner has many functions where, whereas an MRI scanner has many functions, whereas an MEG… Liina Pylkkänen 27:23 It is not actually more expensive. But yeah, I don't want to speculate about all the reasons why this situation might hold. Stephen Wilson 27:32 Okay. And you mentioned like, it's kind of geared towards more, like superficial sources are going to be easier to track than than deep because of you measuring with on the scalp, obviously. Liina Pylkkänen 27:45 To an extent, yeah. Stephen Wilson 27:46 Okay. And then like sulci versus gyri, does it kind of have like anatomical biases there? Liina Pylkkänen 27:52 A little bit, yeah. So it really loves the sulci because the current is kind of oriented in the right way for the MEG sensors to capture the magnetic flux in a really nice way. MEG may have a little bit of a blind spot at the very top of the gyrus. But then as soon as there's a little bit of curvature, the, you know, the fields start to be picked up again. But you know, like when people have really kind of rigorously measured exactly how much signal might be lost, it's it's like so little that it probably doesn't really make much of a practical difference. Stephen Wilson 28:31 Okay. Well, all the important language regions are in the Superior Temporal Sulcus anyway, so that's it. (Laughter) Okay. And so you're using MEG, and you're trying to dissociate, what you call semantic association and semantic composition. Can you explain those two concepts? Liina Pylkkänen 28:50 Yeah, so, so I was interested in this contrasts for quite a long time. Mostly because of my, half of my appointment is in psychology and half of my appointment is like in linguistics, it's like a fully joint 50/50 appointment. And so, you know, especially in my, you know, early assistant professor years, I, you know, I interacted with my memory colleagues, and kind of through that, I learned that there are studies in the memory literature that oftentimes look a lot like our studies. But what they called those studies was all studies on associative encoding, so just kind of building associations between words, as opposed to actually combining them semantically or conceptually. And oftentimes, when they would look at our studies, they'd be like, well, that's kind of like just association. But of course, you know, as a linguist and language scientists, we do not think about that composition as association. They're very different things. So, you know, like, the example that I often use is that, you know, like you're at the breakfast table, you can have a coffee and a cake there. You know, so those representations are very strongly associated in your mind. Right? But that's very different from a coffee cake. Stephen Wilson 30:19 Right. Liina Pylkkänen 30:20 So, two representations can just be associated. But that's different from actually combining them together into, you know, a conceptual combination. And also, also, in this case, a compound or, or linguistic phrase. Interesting thing is, and so this was really what got me interested in this question, is that, you know, so we very consistently see effects of composition, and the left anterior temporal lobe. And we've been able to show that really is more of a conceptual process than a process where you're taking like syntactic categories and building a phrase. And, but the surprising thing is, or maybe not surprising, depending on where you're coming from, is that associative encoding also drives the left ATL, including in the monkey brain. It's very interesting to think about that contrast, and the, you know, possible connection. So, you know, I think it's imaginable that our kind of combinatorial ability might stem from some more elementary associated encoding, like there could be a real relationship there. And the kind of spatial proximity was very intriguing. So, so I had been interested in really like kind of looking into that contrast in a controlled fashion for a while. And you know, oftentimes, when I teach my graduate like advanced seminars, for PhD students in neurolinguistics, I sort of float experimental ideas just to see if anyone bites or wants to pursue them. So I had liked this idea of playing with the associations between countries and foods. That's clearly something that you know, it's kind of easy to vary. And then the kind of cute thing in English is that you can usually turn a country name into an adjective with a relatively small morphological change. So, like ‘Italy’, ‘Italian’, you know, ‘Korea’, ‘Korean’. So, you could create pairs that are just either associated or not like ‘Korean kimchi’, ‘Korean tacos’, and you can vary the, the phrasal structure, so you can just have ‘Korea’, ‘kimchi’, or ‘Korean kimchi’. This is a pretty minimal, minimal change. And so then with Jixing, who already had done some work on kind of composition versus association, so she had some prior fMRI results showing some sensitivity in the left ATL to just level of semantic association in a naturalistic text. And so then, you know, our interests kind of converge. And so then we did this more controlled study. So it's like a very basic crossing of composition where we're varying whether the first element is a noun or an adjective. So the noun doesn't really go with a subsequent now and as a phrase, whereas the adjective does, and then the level of association. So, you would have ,Korean Kimchi’, ‘Italian Kimchi’, ‘Korea’, ‘Kimchi’, ‘Italy’, ‘Kimchi’. Then, you just get a picture afterwards that you have to match to it. And we wanted to see whether the association and composition would have sort of a similar effect and the left ATL. Stephen Wilson 34:02 Okay, so I'm just gonna restate the design of the study just because it's so important using the examples you have in the paper, so you were told about France and Korea and cheese in this case, in the in the, in the figure in the paper. So, you get this two by two design. Things can either be composed, so it'd be like in that one that'd be ‘French cheese’ or ‘Korean cheese’ or they could be not composed or just lists, which would be ‘France’, ‘cheese’, or ‘Korea’, ‘cheese’. And then you're also varying the extent of association. So, ‘French cheese’ is much more associated than ‘Korean cheese’. And that should apply to the list version as well. Right? So that's, that's the two by two crossover design. Liina Pylkkänen 34:51 Yeah, very simple. Stephen Wilson 34:53 Yeah, it is but it's, it's like elegant. So then, you present these to your participants and record their brains with MEG (Meg) or MEG (M.E.G.) as you like it. And I kind of wanted to like in terms of the results, I wanted to start with one of the simpler ones. That's not really the main point of the paper just kind of shows that it's all working, which is the frequency effects that you observed. Can you, I don't know if do you, It's a couple of years ago, do you remember how it comes out? Liina Pylkkänen 35:32 I remember there are frequency effects. But ya know, this is definitely very prepandemic work. So I sort of remember the forest, but maybe not all the trees. I assume the frequency effect is like the usual kind. So temporal cortex. Stephen Wilson 35:50 Yeah. Liina Pylkkänen 35:52 Yeah. Stephen Wilson 35:52 Yeah, you found this middle temporal frequency effect. And it's really nice in the paper, because you see the timing of it, right, like, so you've got these two words, these two word combinations. And like, you know, exactly, you know, just afterward one is presented, you have this frequency effect, for word one, and then, and then just afterward two is presented here frequency effect for word two. And it's really neat, because like, you can just kind of see that, like the brain activity, like reflects the frequency of the word with less activity for more frequent words. Right? Liina Pylkkänen 35:52 Okay. Cool. I should include that when I talk about this study which I haven’t done. (Laughter) Stephen Wilson 36:31 I don't know, it was just my way, it was just my way of making sense of it. Because I wanted to start with, I mean, to me, I wanted to start with a manipulation, where I had a hypothesis about what it should look like in the brain. And to me, that's the manipulation that I have prior. And so I, I like to start with that one, because it shows you that, you know, it makes sense. But then that's not really what it's all about. Right? So, you want to talk about like the main findings regarding that two by two design? Liina Pylkkänen 37:08 Yeah, so this study, obviously, one half of it, is kind of like your totally ordinary priming study. Right? So we have highly related pairs, and lowly related pairs. So ‘France’, ‘cheese’, ‘Korea’, ‘cheese’. So in some sense, that to me, is kind of sanity checking. So, we have massive literature that kind of tells you what direction that effect should go. So we should see some sort of a amplitude reduction for the highly related pair, right? ‘France’, ‘cheese’, everyone predicts like, if you have any kind of response time tasks, that should be faster. We you know, in the electrophysiological responses, there should be a lower amplitude, usually we see it in the N400. But that's kind of a very, you know, there's lots of prior data on that kind of contrast. Okay. And so now, although so far, in this chat, we have been talking about the left ATL, just in terms of its sensitivity to composition, it's very well known that it also just responds to single word, so no one's claiming that it's some kind of composition selected region. There's lots of compelling evidence that it definitely represents, you know, just single word representatives with as well or does something with them. And so now, what's interesting here is that when it comes to that, what we call the lists stimuli, where we just have the two nouns that they're not forming a phrase, we see exactly that type of effects. It's like an amplitude reduction for the more related items. So that's ‘cheese’ when it's been preceded by ‘France’ as opposed to ‘Korea’. Okay, great. Stephen Wilson 38:48 Okay, so ‘France’, ‘cheese’ has lower signal in the ATL than ‘Korea’, ‘cheese’? Liina Pylkkänen 38:53 Yeah. But the the main, the, the result that we report is that, that pattern is an interaction. So when you put the phrasal context into it, so now we have ‘French cheese’ versus ‘Korean cheese’, the effect actually flips. So, ‘French cheese’ is actually elicits a lower left ATL amplitude than ‘Korean cheese’. So, actually, when it comes to those combinatorial representations, the left ATL, you know, activates more for that kind of higher associative pair in the presence of the phrasal structure. So that's kind of you know, that's very intriguing. Stephen Wilson 39:39 It sure is. Liina Pylkkänen 39:40 That we have managed to take, get this kind of like antipriming effect by adding just a very small change, just changing the category of the first element. So, I just think that is really interesting, you know, I mean, we obviously have to understand it, you know, better by running future studies, but it I do think it’s very intriguing, that you know, like the and that we did not predict this at all, like we mainly were interested in eliciting the the effect of association and the effect of composition and we kind of wanted to compare them spatially even though our technique is not awesome for a spatial resolution, but we still want to see like, is there some kind of dissociation? Or do they really affect the same kind of energy source, sources and this interaction was a surprise. Like okay! Stephen Wilson 39:53 Yeah, because that's like the first thing that you report in the results, which sort of, you know, follows classic, like, you know, psychological reporting practices, right? If you've got a two by two design, you start with the interaction, because you don't want to be like messing around with main effects if there's like an interaction, right? So you start with this interaction, and it's rather shocking to see that the, yeah, so why do you think that is? Why would the ATL show an enhanced signal for ‘French cheese’ over ‘Korean cheese’ when it's showing the exact opposite if it's, you know, just noun-noun lists. Liina Pylkkänen 41:10 So, I don't know, the, you know, there's a few different ways that you can sort of begin to think about the hypothesis space, I think one possibility could be that it has something to do with the fact that it is a highly associated, I mean, it's basically there is an existing representation for ‘French cheese’, obviously. And there isn't really for ‘Korean cheese’ and so then when you're putting together ‘French cheese’, you are getting that kind of like activity, boost from, you know, presumably some kind of a combinatory process. And maybe what you're adding on top of that is sort of an activation of that existing memory representation as well. And maybe when you just have the list pair, you know, ‘France’, ‘cheese’, that combined concept is not activated in the same way. This is one, I think, like direction that one could sort of think about that there, you know, there's sort of an added activation of a of an existing representation in the phrasal case when you're actually putting together the concept. Stephen Wilson 42:32 Yeah, I think that makes sense. That was how I was sort of thinking about it, too, as I was reading it. You have this, I'm sort of skipping about, I guess, like, I'm presenting your paper in a different order than how you wrote it. But you know, you have the second analysis, you do where you use searchlight multivariate pattern analysis. And that kind of has this really interesting finding in the ATL as well, where you find that the left ATL distinguishes very, like the this sort of pattern recognition algorithm can distinguish very well between high association composition, ie ‘French cheese’, and low association composition, i.e., ‘Korean cheese’, but it can't really do any of the other classifications. And I guess that kind of bolsters what you just said, like there might be something special about these sort of almost lexicalized combinations. I mean, do you? I guess, I haven't seen the rest of your stimuli. I mean, do you think are all of your stimuli as evocative as ‘French cheese’? Liina Pylkkänen 43:35 I think we have kind of a visualization of the full stimulus, um if, they're not, you know, they obviously vary to some extent, but we, you know, we do have ‘Mexican tacos’. And so one thing that is kind of interesting is that for a lot of these cases, the composition is kind of trivial. You know, like Kimchi, well, Kimchi is Korean, you know, like the modifier is, in some sense, always, almost like semantically redundant for many of the pairings. And especially in light of what we knew about the left ATL from our prior studies on the effects of conceptual specificity on this activity, where we had shown that if you have a conceptually more specific context item, integrating to the current item, you get a larger composition effect than if you have kind of a vaguer item integrating into the current item. So if I have for example, ‘tomato soup’, I'm measuring activity on ‘soup’. So the impact integrating ‘tomato’ versus ‘vegetable’ soup, ‘tomato soup’ will give me a higher effect. Okay, so like, it's kind of more I'm putting more information into the noun. So now from that perspective, it's very weird that ‘French cheese’ would give me more activity than ‘Korean cheese’ because like ‘Korean cheese’ is like, whoa, you know, that should be very surprising, very informative, whereas French, much less so. So that I find that, that contrast very interesting. So I do think a pretty big proportion, I'd have to like, look at them more carefully here, but yeah, I think a lot of them were this kind of like ‘Italian pizza’, you know, like very strongly associated terms, but not, not like expressions that we maybe say that much because you know… Stephen Wilson 45:36 Yeah, ok, you don't go around saying, oh, let's go and try some Korean kimchi tonight. (Laughter) Liina Pylkkänen 45:43 Um, yeah. So. Stephen Wilson 45:46 So yeah. Okay, so you contrast this ATL finding with different patterns in the more posterior temporal regions, like what you call middle temporal. You have an effective composition in the middle, sort of mid MTG and STG regions? Liina Pylkkänen 46:05 Yeah, so that, you know, so the main effect of composition localized more posteriorly and so, you know, and, like, given the state of the art today, I think we're starting to have kind of a convergence of findings that the more posterior temporal cortex, you know, may be kind of a more structural site. And, you know, from our lab, we've, you know, published, you know, four or five papers, kind of, you know, implicating this as well. So this finding is certainly consistent with that. So there, the association level didn't matter, just whether it is actually a phrase or not. Stephen Wilson 46:45 Yeah, I totally agree with you about the more posterior temporal area has been more structural and in contrast to the more anterior being more conceptual. And there's, I think there's lots of lines of evidence for that. I think the direction of your finding here was surprising, though, right? So you had more activity in the list condition, regardless of whether they were closely associated or not. And less activity in the compositional condition. Was that an unexpected direction with the effect? Liina Pylkkänen 47:11 Yeah, so um, so the tricky thing in our hands about the more posterior, more posterior temporal cortex and its structure sensitivity, is that the facts that we've seen just have not been as consistent as the ones that we see more anteriorly. So we definitely see sensitivity there to you know, like, keep the conceptual structure constant, just very structure. But then exactly the directionality of the facts, the exact timing just hasn't been nearly as consistent as for the left ATL. So for that reason, you know, if I give, like a full length, talk about this topic, I started tend to say that the left ATL activity is something that we can kind of hold and examine. But the more posterior effects are just much more slippery in our hands, and much harder to kind of gets a, get a handle on that. So I don't understand that the more posterior, you know like, we have not been able to kind of make mechanistic progress on the more structural effects. Stephen Wilson 47:12 Yeah, that's really interesting. Liina Pylkkänen 48:21 I wave my hands when it comes to that but something’s going on there. (Laughter) Stephen Wilson 48:28 Yeah, the nature of the effect makes sense. That yeah, I mean, I guess there's any number of possible explanations. Liina Pylkkänen 48:35 Yeah. I mean composition of course, is, is if facilitatory factor in language processing. So, you know, we, you know, when it was, you know, when you have a coherent sentence or coherent structure, we remember it, well, it's easier to process than just lists that don't go together. So there's always that way of thinking about, you know, a possible reduction in amplitude as a function of, you know, the presence of syntax, but then, you know, what, that exactly reflects is harder. Stephen Wilson 49:07 Yeah. So, do you feel like this paper was effective at this separating out the different neural processes involved in association and composition? Liina Pylkkänen 49:20 Well, it definitely is a step forward. So I, you know, the research question that we had, was answered. So, you know, and so, you know, like, if you are trained in my lab, the way I always, you know, sort of explain the endeavor of the scientists is that the beginning of the exercise, all these hypotheses on are on the table and your task is to try to get hypotheses off the table. And so here, you know, so one of the hypotheses that we had was that, you know, the, the composition and association would kind of have a You know, a similar effect, you know, maybe in a, you know, a similar brain region. But that that was not the case, that's, I would say, off the table. So instead, you know, the effect of association really depends on whether you actually aren't combining or not. So I think that allows you to sort of formulate the next set of questions. Stephen Wilson 50:25 Yeah, I'm sure that's what you're doing. (Laughter) And kind of as a big picture sort of thing, you know, you've done a lot of very controlled experiments like this, we were you, you know, make these sort of, you know, petri dish kind of situations where you carefully manipulate just one aspect of linguistic structure. You've also done naturalistic experiment where you've looked at narratives and, you know, kind of done pausing of the narratives and looked at brain responses to that. So I was wondering if you can sort of talk about, like, what you see is the pros and cons of those two different approaches that both of which you followed, but I think you've more taken the line of these kinds of like, very fine grain experiments like the one we're talking about. Liina Pylkkänen 51:12 Yeah, it really depends on what kinds of students are in the lab at any time. So you know, at any point, if I had a student who was really wanting to go back to a more naturalistic method, I would happily do that. I'm doing that. But, um, but yeah, I mean, we clearly need both. Um, so the naturalistic methods obviously, tell us something of what happens in more ecologically valid circumstances. They're really good for these kind of large scale model comparisons, like, okay, which model wins in explaining brain data doesn't mean that that's the right model, but at least we are, you know, we can sort of forget about the other model with the other, if one of the models one. So that's all important, an important part of the field. It do feel that the kinds of questions that I really have the most curiosity about are hard to address with a naturalistic method. So you know, because I, oftentimes we sort of take advantage of specific types of linguistic phenomena that like, give us a, you know, an angle at a kind of a cognitive neuroscience question like, that is the combination that I find really fun. And that usually then, you know, invites a more controlled experiment. So they, you know, both of them have a role, I think I personally tend to feel like I'm learning more from the controlled experiments, but that's just like, it's kind of more how my brain works. And, you know, I just, yeah, love a good experimental design, like, that's really fun. Stephen Wilson 52:56 What do you think are the biggest challenges when you when you're doing the, interpretation, when you having people do a controlled experiment, like, you know, just, you know, to is the processing of two word phrases sufficiently analogous to, you know, more ecologically valid, you know, syntactic and semantic combinatorics? Yeah, that's my question. Liina Pylkkänen 53:23 Yeah. So we, of course, haven't just published these two word phrases, or sometimes people think we do but you know, so we have looked at the left ATL, kind of conceptual combination, in fact, in several, like full sentence, studies, as well, not naturalistic, but like the usual type of sentence processing. And, you know, there can be some complications, but we definitely have demonstrations, that you get the similar types of effects and a full sentential comp. Context. Like for example, if you have a, you know, a sentence, final predicate, and then you have kind of a complex subject that has, maybe you're varying conceptual specificity in the object, toggle position, and then you're also messing around with the polarity of the determiner, like, you might have a negative determiner, ‘no, no angry lizard’ versus ‘an angry lizard’, and then, you know, sleeping, all of that is calculated into conceptual specificity fact that the end of the sentence in a way that you would predict from the minimal composition. But there's many ways, that in many ways in which you can model the picture, which obviously, like if you have a naturalistic stimulus, it just gets really complicated and it's really hard to model everything that might go into it. So, but that's, that is, that's the exercise like I do think it's useful to try to discover what happens in a really minimal tractable situation and then you can complicate, complicate that situation and see what changes. But if you never found out what happened in the situation, then, you know, I do feel like there's something. Stephen Wilson 55:12 No, I agree, I think it's important to look at it at all levels. And I mean, and you've and you have looked at it at all levels, right? I mean, you've done minimal combinations, and then control sentences and then all the way up to narratives. I think that we want to look for these different ways to kind of inform one another and paint the whole picture. I remember when Steve Pinker was, had decided that like he was only going to study the English past tense, and that was basically like the fruit fly of language science. Liina Pylkkänen 55:44 I remember. (Laughter) Stephen Wilson 55:47 Yeah. So let's kind of just talk a little bit to wrap up about SNL. So did you enjoy our first in person conference? And we just went to in Philly? Liina Pylkkänen 56:01 Yeah, I absolutely loved it. I thought it was really, really great meeting was so nice to see everybody. I was really happy with the attendance. Like I felt like that room was full pretty much in every session. Like I felt like everyone was going to everything, which is not, you know, normally the case. Like there's some sessions that aren't as popular as others. But you know, I was always kind of like, looking around. I was like, Oh, this people really consistently showing up. So I thought it was really, really great. I really had a great time. How about you? Stephen Wilson 56:30 Oh, absolutely. I mean, I've missed it so much, you know, just catching up with everybody. Liina Pylkkänen 56:35 Yeah. Stephen Wilson 56:35 It was really nice to be back in person. Were there any like particular highlights of the conference for you like posters or talks that that just kind of really captured your attention? Liina Pylkkänen 56:46 I loved actually, this is a joint student of ours, Deborah Levy. So her poster was super cool. Were you a co-author on it? Stephen Wilson 56:58 I, yes. But I didn't do much on it. Liina Pylkkänen 57:00 No, because we did a lap meeting, which is like SNL souvenirs, and everyone sort of shared their favorite one. So that's the one that I shared. So you know, I was really excited to see it. So this was from Eddie Chang's group. So she's now a postdoc there. Stephen Wilson 57:18 That's right. Liina Pylkkänen 57:18 Your grad student, but she was an undergrad at NYU. And so whenever poster, she actually told me that she had taken my Neural basis of language classes. And so, so, you know, so they had looked at language performance, after temporal lobe resection and epilepsy surgery, you know, asking, what resections cause Wernicke’s aphasia, and it was pretty interesting for me to find out that those resections were very left ATL centered. Stephen Wilson 57:55 Yeah. Liina Pylkkänen 57:56 Um, and so you know, because the neuropsychological profile for the ATL does create somewhat of a tension between, like our findings with with our findings, and so, you know, like the usual profile that one reads about, like, I'm not a clinical person, but you know, it's more of a kind of a single concept deficit, if you have atrophy there, you kind of lose the more specific vocabulary, but no one talks about like composition or sentence processing. So it was really interesting for me to learn that, you know, immediately after resection, you do have, um, you know, like a Wernicke’s profile that they then recover from. But anyway, that was… Stephen Wilson 58:44 Yeah. Liina Pylkkänen 58:45 That was what I was like, really like… Stephen Wilson 58:49 I'm really glad to hear that you liked that one. I liked it too. Of course. Yeah. Cool. So, you know, you're now the president of SNL. And what kind of things do you think that the society can do? Like what do you hope to do as President to make this society you know, better able to support the field and support all of us that work in this field, you have plans for the year, Liina Pylkkänen 59:17 I do kind of feel the challenge. So we are now in this, you know, place in the world where we have to kind of like invent the future of conferencing. You know, we somehow have to have a more climate friendly approach. And now we have learned how you can have tremendous access across the world by doing all sorts of virtual things. So we rapidly learn that so we see all the pros and cons like you know, most of us are not huge fans of online conferences. I certainly never really got into them. But at the same time, we have to recognize that we have to sort of try to get more global, global access. And so, you know, I think the, you know, the concept that we'll, we're board buddies now, since you're also on the board. Stephen Wilson 1:00:15 Yeah. Liina Pylkkänen 1:00:19 You know, I, you know, I think we'll, we'll discuss this idea of having more virtual events or some kind of virtual activity through the year, while still being able to have an in person conference, which, you know, I do, and I know how you feel about it is really, I think, important. So just, you know, I do, it's tough. Right? And so hopefully, we can make a little bit of progress this year, trying to, you know, settle into some kind of new normal. Whatever that is. Stephen Wilson 1:00:52 Yeah, I really liked the idea of making it additive, right? So like, not taking anything away from the in person conferences, but just adding something new as well. Right? Adding virtual interactions and, and virtual opportunities, because that seems like the, the best of both worlds potentially. Liina Pylkkänen 1:01:09 Yeah. Stephen Wilson 1:01:10 And so yeah, Marseille in 2023, we're gonna be there. Did you go there recently? Are those were those your own pictures that you posted? Liina Pylkkänen 1:01:18 Yeah, I did actually go there. Last summer, I had some other reasons to go to Marseille as well, I actually have a PhD student there who I hadn't yet met in person. And so I was excited to see her. And then I also checked out the conference venue. I really wanted to see it because it's a good feel felt weird to me to be organizing this conference there. And I had never been there before. So you know, I was already in Europe anyway. So I got really excited because it's a really beautiful city. And the venue is really, really beautiful. So I think we'll have a great time. Stephen Wilson 1:01:55 Yeah, I can't wait. And do you think we should change anything up in the program? Like people are always asking me like, when are we going to bring the debates back? I mean, we should like, bring the debates back, or have you got any other thoughts about like programming? Liina Pylkkänen 1:02:08 Debates? That's an interesting question. Yeah. Those are fun to watch but then you have to find the debater. Yeah, so I don't know. I don't know if that concept will fly in the in the new world. Stephen Wilson 1:02:22 We can just make it, Greg Hickok versus everybody. He’ll be into that. (Laughter) Liina Pylkkänen 1:02:26 But you're the chair of the program committee. So you're the one who has to tackle with this question. Stephen Wilson 1:02:31 Yeah, I know, I know. I'm just that's why I'm trying to like (Laughter) ask for other people's ideas. Liina Pylkkänen 1:02:38 I like it when there's lots of time, to just to interact. Stephen Wilson 1:02:42 Yeah. Yeah, the program was very full this year and I think I mean, I don't think unreasonably so, because there was a lot to cram in. But yeah, there were not many gaps. Liina Pylkkänen 1:02:57 Yeah. Okay. Stephen Wilson 1:02:58 We need to leave time for French cheese. Liina Pylkkänen 1:03:02 We need time for dinner. For the evening. Yeah. Stephen Wilson 1:03:08 Yeah. That’s gonna be a lot of fun. Looking forward to working with you and the other board members on it. Liina Pylkkänen 1:03:14 Same. Stephen Wilson 1:03:16 Cool. Well, I'm glad we finally got to do our podcast, after like, half year of trying. Liina Pylkkänen 1:03:23 Yeah, yes, me too. Very good. Stephen Wilson 1:03:25 Yeah. So, thank you, and I will catch up with you soon. Liina Pylkkänen 1:03:32 Thanks, Stephen. Thanks for doing this. You're doing a huge service to the field. So… Stephen Wilson 1:03:36 It’s a lot of fun. Liina Pylkkänen 1:03:37 I really appreciate it. Yeah. Stephen Wilson 1:03:38 It’s a lot of fun. All right. Take care. Bye. Liina Pylkkänen 1:03:44 Bye. Stephen Wilson 1:03:45 Okay. Well, that's it for episode 24. Thank you very much to Liina for joining me today. I've linked to the paper we talked about in the show notes and on the podcast website at langneurosci.org/podcast. I'd like to thank the journal Neurobiology of Language for supporting transcription, and Marcia Petyt for transcribing this episode. See you next time.