 A, I believe we're live, the usual brief wait to make sure that everything reconnects correctly. Zoom even adds a little bit of extra delay making sure that I wait for this to switch over and we'll wait just a second more to get everybody into this session but it looks like everybody is into the session. So cool, all right. Welcome back in everyone to the second to last talk, the last block of the conference. I was just saying as we were getting connected this was actually a sort of serendipitous network block. So we've got our third talk making use of drawing awesome insights from network analysis. This time though, this is another aspect of this work that I'm really excited to have in the conference that is to say another important connection and I think we need to be drawing is with the digital humanities more broadly which is of course about studying more than just science. So now we're gonna get to see a little bit about network analysis as it's applied to poetry and I'm really excited to hear this talk from our very own local Chris Tannis-Eski who I've worked a bunch on a variety of other initiatives and as you can see here, I'm not even gonna read out a number of co-authors both at UC Nuvan here with us, oh, you popped out of full screen mode. Oh, okay, sorry. No, no worries. I just happened to notice it as I pop back over here to the other window, both at UC Nuvan and at the University of Ottawa. So let's see, hang on, let me make sure that we get the video thing sorted. You just need to go up to play, is that simple or are you trying to get like presenter notes and stuff to work? No, I'm trying to get it back to play but I needed to stop the video from the website because there was like this clash, you know? Oh, oh, you had the other window open, I understand. No, I don't want to get back to the, maybe I should stop the screen sharing and then to share again. Yeah, that might be safer, yeah. Yeah, yeah, okay. Yes, the infamous crowdcast feedback echo as we saw it the other day. We don't want that. All right, please, there we go. That looks perfect, take it away. Thanks. Thank you, Charles. So hi everybody, it's good to be here. It really feels good to be part of this amazing, amazing conference, heartiest congrats to the organizers and the presenters. And indeed, what a serendipitous amazing coincidence, quote unquote. Well, I will be like, you know, the humble follower of my predecessors and this meeting. And I'm also just a representative of a team of an international team. Just like Charles said, we have a couple of guys working here at UC Le Bon and a few others at the University of Ottawa in Canada. And this is part of a project titled The Graph Palm. I will give a bit of a background. And then I will focus on some of the most recent, should I say, advancements. We're trying to translate that into advancements on closeness and betweenness, centralities and multiplexes and their application in the poetry. So speaking of the background and the concept of all this whole story started, it goes back over a decade. The concept refers, I mean, I hope the name is pretty self-explanatory, kind of speaks for itself, right? It refers to assembling networks of poems in which the poems are the nodes and the edges are correlations between poems in various respects. And that's where the first difficult is kicked in. But before getting there, one thing at a time, what we adopted from the get-go was an asymptotically holistic approach. Well, asymptotically kind of speaks for itself, of course. Holistic may sound a bit presumptuous, but in our case means pragmatically tackling poetic features, shooting for all poetic features, that's why asymptotically, and then fully treating poetic features. And to the best of our knowledge, we're the only one ongoing project in the English-speaking world, at least, undertaking to do that. While being aware that, as I already suggested, all features is already something that might sound presumptuous, but besides that, poetic features, they are critically debatable, questionable. There are so many takes on meter to begin with, can be defined in so many ways, and there are so many approaches to meter, some of them completely different from the classic, the established, whatever you wanna call it, acceptance of that. But this is something maybe I'll get closer to and deeper into during the Q&A, but we also refer to the door approach as being holistic in other respects as well, and particularly in the sense of the intermediality of poems being embedded in the medium and in a wider context in digital space. Again, something to be detailed if not during the presentation per se later on. And then, of course, now since we speak of features plural, of course, we will speak of multilayer networks, networks in which every single layer will represent a feature, right? And since we speak of corpora of poems, then we would expect, of course, to have the same nose in each layer. So technically speaking, we'll be speaking of, we'll be dealing with multiplexes. Now, for a brief history of this, very brief, but then on a personal level, when I first came up with this idea, I came up with it from the standpoint of a poet myself and the professor of poetry and creative writing. Over a decade ago, I was a postdoctoral fellow at San Diego State in California. Wow, yes, paradise, going swimming, going for a swim, for a surf in late November. And well, it started with the poetic manifesto on legendary Jerome Roddenberg's online journal slash blog, poems and poetics, and then followed by a number of other publications. Like I said, from the standpoint of a practitioner myself, of a poet myself, an academic in poetry and literature, but nevertheless having a background in mathematics and computer science. One of the milestones of that stage was an anthology, a manually assembled graph poem, the first attempt at doing anything palpable, doing something palpable in that direction, over 50 poets from all over the world in various ages as well, under the collective name of Margento. That was the first book in a line of publications putting this concept into practice. Well, not for self-promotion only, and I would hardly speak of self here since it's the whole team and that's why we've been using this umbrella term, Margento, but for the sake of the kind of conversation that we started and the moment we got the ball rolling, the kind of critical reception we got, and this is one of the probably the most significant ones from David Baker, the guy needs no further introduction, significant major contemporary poet himself, also editor of Kenyan Review, and his enthusiastic reaction to this initiative. Well, enthusiasm and worm reception aside, what I can, if I were to be the advocate of the devil here, from this excerpt of what he wrote about Nomodosophy, the first graph poem book coming out of this project, is that he never mentions here and not only in this excerpt that goes for the whole piece, he never mentions the fact that it's a graph theory based project, right? It might be communal, impressive initiative, very good poetry, but where's the graph? Well, there is something still that got in the picture, the performance side of it, and I will be going back to that, but for now I wanna also briefly highlight the fact that I'm really glad that from the very inception, people like David Baker, for instance, noticed the strong performative dimension of the graph poem, and that actually goes very intimately connected with the computational approach and the concept per se. So I would say the performativity is part of the concept, and generally speaking of any approach that would consider applying network science or graph theory in poetry, more about that in a bit. Well, and these are some exemplifications in that direction, so Margento actually took off, even before that, as early as a matter of fact, as 2001, as a performance cross art form digital poetry project, computational poetry would be a better term probably than digital poetry. Again, a discussion that we could have later or forget it for even later on, but I'm just highlighting a couple of events. We actually did such events in Europe, in North America, in Vietnam, in Australia, you name it. I'm just highlighting again, a couple of milestones. One of them was the poetry conference in London in 2013, and then the sequence of performances at DHSI, starting back in 2019. So this year is gonna be the third one, and this is the year 2021, when the director, Ray Simmons, preferred to use the term the institute performance for this event. This is just a screenshot of the announcement and description of the event for the upcoming edition, June 2021. Again, we can delve into details a bit later, and as part of the presentation, I'll have to go back to that in a certain respect. Well, NLP classifiers, NLP, of course, natural language processing, speaking of what Charles was mentioning certain deputy before a certain deputy kicks in. You have this like, when you start, and I'm sure that you guys, I would expect everybody is on the same page with me. We can share such experiences, right? You start the project and then you go like, well, but what's your concept? Well, I want to organize poems in networks and represent poetry corpora as networks, and then use them for creative writing purposes, for teaching purposes, and so on and so forth. And then, of course, it's like, oh, that sounds amazing. How are you going to articulate the networks? Well, of course, we need some connections between the poems. Yes, okay, so what do we need in order to be able to do that? What do we need to do? Well, we need some classifiers. And that's, again, I think you guys are on the same page with me when I say, well, you think of it as like, oh, I got to do this in order to do that. So maybe it's like a preliminary, something like, I'll have to sort out as fast as possible so that I can have fun with my networks. Well, it turns out these preliminary things can go on for over a decade, in this case, like, you know, casing point, and they can actually, now I'm aware they will go, they will go on for a lifetime. But enthusiastically, you know, with an age of the beginner, I was like, oh, okay, let's put this in place. Let's step this up. So I got a grant in Canada, a shirk, like an acronym for social sciences and Humanities Research Council to exactly do that. Well, actually the grant was for like, you know, developing the graph poem, but in order to do that, yes, we need to develop MLP classifier. So classifiers roll on, and that's what we took, we took them one at a time and then we realized we'll have to go back to every single one of them, one at a time, we can. Then we realized, no, I know all of them at the same time, but we need to go back and then, oh, what's there left that we never touched so far? So here, I copy pasted a couple of very short descriptions of work we've done so far. We started with topic and soft topic and this will, of course, inevitably adhere into a discussion on data and, of course, the keynote today, for instance, among, you know, so many other brilliant presentations over this past four days, but the keynote today definitely kicked it out of the ballpark. We all know, again, my guess, well, speaking for me and my team, we definitely know what it means to like, oh, where's the data? Well, speaking of, for instance, this first classifier that we developed, well, Poetry Foundation is the English language poetry online archive that is available for free and is manually annotated rigorously so, and is there, you know, offer grabs, unlike, for instance, the ProQuest one. We did this, then we moved on to meter and rhyme, scratching the surface of the first classifier, focused on, you know, basic meter in English and, of course, when you speak of that, you speak of Iambic pentameter, you'll have to go there and also touch rhyme, first attempts were like, you know, just very basic, two classes only, we had to go back and I will go back to that in just a bit. Well, in the process, interestingly enough, and I know you guys know from experience again, well, you work on something and you, serendipitously again, stumble upon other things, right, or you have like, you know, illuminations, you can have, you know, all these moments of enlightenment, when you go like, oh, so while working on, focused on poetry, it turns out that, oh, I can answer issues in other walks of life and particularly in an LP, and that's what happened to us. Well, Joyce would call it epiphanies, right? Yeah, we have this epiphany. Well, it wasn't that mystical, and in Joyce's case, also not that mystical, it was pretty much mathematically explainable in the sense that, well, who did work in metaphor before? Well, people who didn't deal with an LP, that is, people who didn't deal with literature. Well, when we combine data, we notice that, that it's poetry and non-poetry, and that's what I answered here very briefly. We notice that, oh my God, we can obtain better results than previously, that the results previously obtained by people focusing strictly on non-poetry data. So, but it also works better for us in poetry when we do this mix, very interesting. Well, that kind of bagged, at least as far as I was concerned, bagged for trying to make an argument in favor of poetry's relevance beyond the genre in digital humanities, well, in natural language processing and in a wider context in natural language processing. We took this to the next level, what's the next level? Well, we all know, deep learning. We did that in a metaphor detection, it turned out very well, and that, again, made us stumble upon some more serendipitous, again, at times incredible results. Like, for instance, developing word embeddings that were better than glove, well, who doesn't know about glove, right? Well, just training word embeddings on poetry data amazingly so, output it better results than training word embeddings on the usual kind of data, the news data, we all know about that, the, all those huge databases and corpora that basically never includes, well, let alone poetry, but in literature in general. Unless you want to consider a very permissive and I'm not against that, and that would be the first one, definitely not the first one to accept that, but the point is, when you focus on data that can be, quote unquote, universally acceptable, considered to be poetry, and then you don't find it in data that is usually used, that is usually fed to word embeddings, and then you mix the two, you train the word embeddings on data that is labeled in terms of metadata and all that poetry, and the word embeddings are better, well, you definitely have a shot at making an argument in favor of poetry's relevance in digital humanities, I would say. Well, next step was we developed a black box. Well, all of these classifiers, and I only mentioned a couple of them, diction is also there, then we kept refining the rhyme, rhyme, well, actually the more permissive, the more generous term would be euphony, of course, seven types of rhyme, and we got a forthcoming publication on that as well. We decided, well, we need to put all these things together and make them available for the community to use, and we're in the process of making that available on the URWNLP lab server. One of the co-authors of the paper, Diana Inkpan, a big name in the world class name in NLP, she's the director of the director of VNLP lab at URWN, which is not like saying, well, she picked up the phone and called somebody, and our tools will be up on the server, no, it's not that easy. You guys, again, I guess know about this, is the IT department, is the technical support staff, and all that. Well, we are in the process of, okay, so stay tuned, stay tuned. Well, that's just the screenshot of the rhyme classifier is run on local host. And we have, you can see here like the seven categories. You can get, well, we fed, you know, the Poetry Foundation, again, archive of over 40,000 poems, and you can also click on two or three of them if you also like to have the visualization and this is like the illustration of clicking through dimensional visualization of rhyme classification of a sub-corpus of the corpus. But the classifier nevertheless works on all seven categories. Finally, the graphs, right? I remember Diana, Diana Inkpen, Michael Applicant, University of Ottawa, the co-author of this presentation, of this paper. At a certain point, she was telling one of the graduate students that we were supervising was graduate student who was working on this. And maybe we should finally get to doing some graphs since the name of our grant is the graph phone. Yes. Well, we got to the graphs rather later than sooner. And I'm not yet sure we really got there, but, you know, it's in the name at least. One of the first significant publications that factually used representing these corporates graphs as networks, right? Was a computationally assembled anthology, US poets, foreign poets back in 2018. I'll explain the quotation marks that you see there wrapping the US in the title. Well, to begin with, and very briefly, what it is that computationally assembled means in the title, it means a number of things. Well, first off, as we can see here, the translation of a US poetry corpus into network graphs, that is the representation of this corpus. Well, the quotation marks, the quote on quote there, already kicking at this level, US poetry, well, yes, but no. I mean, yeah, well, we crossed the border of digital and page-based, you know, established even if unfortunate terms, right? Page-based versus digital poetry. We can talk more about this during the Q&A. We crossed that border, but we also crossed the border of US in the initial corpus, even if most of the poets were US poets. We had the exceptions like David Jave, Johnston, you know, Jave, huge name in Canadian digital poetry. Then John Caley, well, American, American academic, right? Teaching at Brown, but also Canadian. And some would say one of the, if not the most important contemporary digital poets. Practitioner and theorist, but also people like Maria Mencia, who's an academic at Kingston University, London, but actually a Spaniard, and so on and so forth. I will again, continue with what competition assembled meant in this particular case, and then go back to the quote on quote thing around US in the title. The first one I said were translating or representing the initial corpus as a graph. The second one was assemblages that would integrate sub-genres like they already mentioned, mentioned traditional slash page poetry and digital poetry or the more general genre, electronic literature, right? Feel it. Thirdly, we did translations of algorithms that generated the originals into algorithms for composing, generating or assembling the translations, translations. Well, some of the, some of these algorithmic translations in the sense of translations of algorithms were like accurately looking for equivalents. But others were like, you know, sticking in stuff. Like for instance, I mentioned Maria Mencia, her poem contributed to the anthology, the poem that's across the Atlantic. We took the, it's an intermediate poetry, poem, sorry. We just grabbed the prose description and wrote an algorithm to translate that into a double meso-stick in which the pillars, you know, the axis of the meso-stick would be reading Mencia and Neruda. Cause, you know, Pablo Neruda was a poem, was a poet that appeared significantly and in her, it appears significantly in, in her digital poem. And also her grandfather who was a refugee coming from Spain and then France to South America. So that was like, you know, sticking in an algorithm that would make sense of the poem, but like, you know, kind of taking it away from, completely from its initial, its initial location while other algorithms would try to like replicate one way or another originals into translations. And fourthly, and perhaps most importantly for our discussion, we developed algorithms that would automatically expand the corpus. So the anthology itself was actually an anthology in progress in the sense that there were algorithms outputting poems to be included on a ongoing basis, on a permanent basis. And that opened the gates of the US in the title to like basically everybody and any kind, in every kind of poetry, everybody, anybody, any kind of poems and of poetry's beyond any borders or boundaries. And that's where the US turned into us, you know, us poets everywhere and anywhere and in any other age, including a certain point, Babylonian poets and their poems in English translation. Christopher Pancauser wrote about this anthology, again, not necessarily for a self-promotion here, but for an aspect that he pointed out, that's why Justin excerpt here and not the whole acolyte, the one that underlines the processual quality of the anthology and the way in which it keeps generating content out of itself and by searching automatically through other databases and archives. Like I just mentioned. And on top of that, the performative aspect of what we've tried to do there. That feeds, again, into the topic of performance. And I hope I'll get the chance to look into what performance really means here besides the actual and the established meaning of the term, which is not to say that we're not doing performance in the literal sense, but the concept that kind of underlies the graph poem initiative is sort of translated acceptance of performance itself while covering performance in the most literal sense of the word as well. And here we see a couple of screenshots of what happened in 2019 at DHSI. So there was a Jupyter Hub with some coding script going on, participants having the possibility to run the code but also contribute to the corpus, the initial corpus that we put together, we, Martento, as the DHSI poem corpus. Everybody had the opportunity to contribute to the corpus and run the code on Jupyter Hub. And the algorithm picked certain particular nodes out of the resulting networks and off of it onto a bot on Twitter, tweeting excerpts of those poems or texts and pairing them up with creative work coming from our general videos, audio files, et cetera, which was kind of a surprise because the event was announced as a performance. And again, I'm playing with this term, but it turned out, well, and people were not weren't that what happened to what they contributed to that shared data and what the algorithm did was being fed into a bot that, and that was the performance. Well, for the 2020 edition, while everybody knew, oh, so the performance is like in a computational way of an algorithmic way of feeding some corpus into the bot, into a bot. Well, the surprise was that it was all the whole thing was also fed into a live stream on Facebook. And we can, well, go into details about that as well later on. Yes, Charles. I just wanted to ping you to say about 10 minutes left in the session, just wanted to make sure we had time for some discussion. So yeah. Yeah. Thank you. Well, destination multiplexes, first stops and trawleries. Multiplexes, like I said, multi-layer networks, same nodes in every single layer, but different layers. In our case, layers representing poetic features. This is an excerpt from the anthology, the computational assembled anthology I was mentioning and another instance of serendipity. Why centralities? Well, we are aware that, you know, you'll have pundits tell you that, well, centralities are maybe significant, but they don't really represent the whole network, right? They will tell you things about certain specific, spectacularly specific and prominent nodes, but how about the whole corpus? Well, that's what happened to us while working on. So it wasn't there in the beginning, this concept, or rather this notion of focusing on centralities. But what we notice and I highlighted here was that certain poems in the initial corpus, the US corpus were very low in closeness and very high in betweenness. Well, in lay terms, and I'm a layman myself, poems that are really, really marginal were actually incredibly effective connectors. And we're like, wow, that's amazing. How about we look deeper into this? And then what we, the kind of algorithm that we developed was actually aiming at this, at cracking down and including in the expanded, in the ever expanding corpus, poems that would act out the same way, be marginal, but also very strong connectors. Well, now what we did in developing that for the anthology that I mentioned, was a single layer network. And now we're moving on to multiplexes, a multilayer network, a particular case of multilayer network, of multilayer network. And the way in which we developed the algorithm that expanded the corpus was, corpus was based off of certain features of prediction of poems. Since we had a diction classifier already and developed in house one, we were able to do that going through databases and archives at hand. And again, there's another tie in with keynote today and not only exactly where were those resources and those archives and databases. And we selected poems to be added on top of the already existing ones and that got the ball rolling at the snowball effect in that anthology. But now we were like, man, we got all these other classifiers, right? So now we had the means to the end, but we don't know how to deal with the end. In the beginning, the NLP classifiers were like, oh, the preliminary stuff to finally get to the graphs. Now we got to the graphs, but we're not able to deal with the graphs. Why? Because the literature available in computer science, and I'm not talking mathematics. Mathematics over the past couple of decades and particularly in the past 15 years, when skyrocketing in terms of detailing concepts and implications of multilayer networks, but computer science, a bit lagging behind. So we're like, we would need to do the multilayer network, but we also need for our graph poems, but we need to contribute to applications of multilayer networks in computer science as well, because there are no effective, no actual tools to do that. And here is a description of what we had to do, computationally speaking, in order to step that up. Basically, we had the component that would plug in the similarities between poems in different layers, and then we had something that we could import, the network acts library in Python that is usually, and you know, establishedly used for single layer networks, but we kind of managed to import it and like repurpose it for multiplexes. And here I'm detailing the ways in which we moved on, like the very first step from a single layer network to a two-layer network. And this second layer was the rhyme layer, and I was like, you know, I was like rooting for that since, you know, the first approach looked into only into diction, and I was like, I was desperate for something that would account for formal features, right? And euphony is one of those, right? That would account for like line breaks and so many other aspects and stresses and so on and so forth on top of diction, which could be, you know, as pertinent or as relevant to other genres as well. I, we mentioned here, there's also, there's actually something that was copy-pasted in here from Nikola, Nikola Burney. Some of the upsides refer to the low coupling and the modularity. We thought of this from the, of it, this way from the very beginning, we needed for it to be very modular because if we needed it for like, you know, several layers of poetic features, then we were aware other people who like to use it might use it for, might need it to begin with for so different purposes and for such different kinds of data. Well, for the, well, speaking of upsides, now downsides of course, well, for the time being is just, you know, a couple of folders on our Macs that we shared between us and that's the way it goes, like, that's the way it goes, right? And, you know, issues like, you know, the Ryan classifier was written in Java. Well, it took like, it will, it would take us for, like, for ages translated into Python, but we wrote an adapter so that we were able to tap into the Java classifier that was written by one of the authors of this paper, Vaibhav Kesarwani. We wrote the adapter to tap into that and then plug it back in into our framework in Python. Well, it's just a screenshot of how it works on VS Studio. I usually use Jupyter notebooks, you know, for a number of reasons that I don't want to go into right now, but, you know, speaking of things that can go crazy, this particular case just simply didn't work on Jupyter and it did work on VS Studio. Well, this is part of, again, maybe for the Q and A of the way in which we look at performance in digital space and what we churned in previous publication, A, poetic technologies, right? The networks go so much deeper than the poem corpora, right? And the networked routines and ways of writing are there basically everywhere in the medium and in digital space more widely. And this is one of the examples of that and who can possibly know for real why and how? Well, it works on VS Code. Well, a couple of closing notes, if you will, on this, from a, I wouldn't call it a philosophical, particularly not in this context of this conference, I'm no philosopher, but, you know, kind of, you know, speculative considerations of what's happening here with this project and these kinds of transitions from, you know, a single layer to multiplexes from a certain corpus to an expanded corpus, what happens to, from certain kind of data that in this approach, you would say the same data, but, you know, as long as we worked with it on and on it on a single layer basis, it was this kind of data and this data and but that or plural, plural, right? These data and then the moment we added another layer, oh my God, these are different data. We can, you know, and the full paper will include like comparisons between the positions of nodes, right? That I mentioned before, what happened to them when we layer, when we added another layer, right? And then if the initial corpus was the US, us actually, was expanded in a certain direction when looking into diction, now that we're looking into diction and sound, are we expanding it the same way? Definitely not. What happened to all these collections of data that are and we developed a quote unquote philosophy on this and we got a publication forthcoming on that as well. All these multiverses of poetic data they go where all these constellations, we look at them as documents of performances of the graph poem as being inscribed in the initial space, multiverses, if we will. These are my references, it's just a brief list. There are so many more. And I'll stop here for now. Thank you so much for your patience and I'm looking forward to your questions and comments. Fantastic, thanks so much. We don't have too much time for Q&A but let's get in, we can. So I wanted to, I'm gonna do my usual and start with one for me while we wait for more people to chime in. Really cool stuff, I'm really excited. I hadn't seen the newest stuff on the classifiers, I don't think, this is really, this is looking really, really cool. One thing that I wanted to grab onto, it's a bit more speculative but I'm gonna go there anyway. Something that seemed to come up at multiple points in your talk was this idea of, and it's actually, it's been a bit of an undercurrent throughout the conference, this idea of these digital analyses as sort of making the unexpected visible, right? Teaching you stuff that you, getting things out that you didn't put in, rendering visible stuff that you never thought you'd be able to see. And I wonder what that, could you tell me briefly what that's meant for you and how you engage with that in your research process? Yes, well, yes. I know how does that, yeah, you're right. It's been like, you know, an undercurrent of this fantastic conference and it's fantastic for this reason as well. The people being honest about this and like, you know, keeping an open mind, right? It's like, oh, I'm omniscient and I knew everything from the beginning. No, for me as a poet is a blessing. It's like, you know, doing experimental poetry. That's the way, and I see that's the way of doing experimental poetry nowadays. Writing algorithms and learning from them things that you never saw before. And, you know, you called it very correctly, serendipity. I also mentioned like, you know, preliminary stuff that gets permanent and then provisional things that become obsessive and so on and so forth. In Romanian, since I'm of Romanian origin, we have this unique term, intemplare. It would be like from Latin, intemplare. It's unique among Romance languages. I mean, like translated, it'd be like what is in the temple? But in the temple, like, you know, when you're doing like, you know, we're trying to foretell the future, but it also looks like haphazard and like totally unexpected, right? And the meaning in Romanian is that, I mean, the etymology sounds very mystical, be it pagan, of course, and why not? But the connotations in Romanian are all that together. It could be providential, but it could be totally random. It could be like serendipitous, but it could be like also absurd, you know, and think of Ioannesca and, you know, the strong absurdist tradition in Romanian letters. So I'm looking at these from all these possible angles. I love it. Let me, let me unfortunately bring it to a wrap there. I'm so sorry, but if there are people who have more questions and comments, please feel free again via the crowdcast chats we're available after. So drop stuff, drop stuff in there and we'll see you guys at the next talk here in a few minutes. Thank you so much. It was fantastically cool. Thank you so much. My pleasure. Thank you.