 We're streaming now. Should be live. Give it just a second. Okay, settings are active. Good, all right. That should do it. We're live, great. Okay, welcome in everyone to the second talk of the day. Without much further ado, continuing a bit our digital humanities methods block. I'm really excited to see this. This is a very cool idea. I'm really interested to see how this unfolds. So this we have with us Eugenio Petrovich from the University of Siena, who will be talking about acknowledgments, co-mention networks as a new method for mapping the social structure of scholarly fields. So please take it away. Hey, thank you very much, Charles, and hello everybody. Thank you for attending this virtual talk. And thanks to Charles and Luca for organizing this very inspirational, let me say, workshop. And today, my contribution focuses on the acknowledgments of academic publications and how they can be used to map the social structure of a science of scholarly fields more generally. So let me start by a brief outline of the presentation. And I'll begin by sketching a schematic picture of what we may call the multi-layer structure of science. And I will briefly overview the main techniques that have been developed in scientific metrics and some science mapping to map the intellectual structure of science. And then I will focus on the main method that is used to map the social structure of science, namely co-authorship analysis. And in particular, I will review some well-known limitations of this method that is based on multi-authored publications. Then I will introduce the acknowledgments of academic publications as a complementary data source that can supplement the analysis of co-authorship. And in particular, I will present you the concept of acknowledgments co-mentioned network or ACM networks in short. Then in the third section of the talk, I will test this new method for mapping the social structure on the era of analytic philosophy that, as you probably know, is one of the main tradition of contemporary philosophy. And we'll present you the methodology I followed for generating the ACM network. And then the results I obtained. And in particular, I will focus on a comparison between the social and intellectual structure of analytic philosophy. Based on the results of this case study, I will discuss a little bit in the fourth part of the talk, the advantages and issues of acknowledgments analysis in ACM network. And lastly, I will conclude by sketching some further steps in this research. Okay, so let's start by what I call the multi-layer structure of science. Science and scholarship more generally are complex multi-dimensional phenomena that can be analyzed from various perspectives. From an analytical point of view, we can first distinguish an intellectual layer of science that is the layer of the epistemic content of science. When we describe the scientific field in terms of research areas, topics or knowledge domains, basically we are focusing on this specific level. However, as sociologists of science and philosophers of scientific practice know well, the intellectual layer does not exist in a social vacuum. In fact, scientific ideas are embedded in a complex social network made of scientists, scientific institutions, grant agencies, and so on. All these diverse actors constitute the social layer of science. Now, in between the intellectual and social layer, a key role is played by the communication system of science. That is the system made of publications and journals by which scientists and scholars communicate and exchange scientific information. Now, the key idea on which scientific and science mappings are based is that from the traces left, so to say, by scientists in their communication system, we can reconstruct both the intellectual layer of science and the underlying social structure. Now, as to the intellectual layer, several techniques have been developed in the last decades by scientometricians and experts in science mapping. The three main techniques are co-citation analysis in which the intellectual similarity between publications is determined based on the number of times they are cited together in the scientific literature. The basic idea behind co-citation analysis is that the more two publications appear together in other articles, the higher the probability that they belong to the same topic. Then we have bibliographic coupling that in a certain sense is the reverse of co-citation. In bibliographic coupling, the intellectual similarity between publication is determined by the overlap in the bibliographies of publications. Again, the idea is that the more references two publications share, the more they are closed in the intellectual content. And lastly, we have co-word analysis. This time, the intellectual structure is reconstructed by mapping the terms that mostly occur together in the titles or abstracts of publications. Now, as to the social structure of science, however, science mapping offers a less rich toolbox. In fact, the main method that has been used that is used is co-authorship analysis. In this type of analysis, the social relationships between researchers are reconstructed based on the publications they write together. The idea is that scientists frequently writing publications together are likely to belong to the same invisible college that is those active scientific groups that form the basic unit of the social structure of science. However, this type of analysis that is based on these formal collaboration patterns expressed by co-authorship suffers from some well-known limitations related to the practice of authorship in science and scholarship. First, it is well known that authors contribute differently to publications. Usually some of the authors are more involved in experiments, analysis, and writing, whereas others play a less prominent role. To account for these differences, some disciplines such as biological sciences rely on the order of authors, whereas some journals use specific taxonomies that allow each author to specify their contribution to research. These innovations such as the credit system may partly solve the issue of waiting the contributions of the authors. Nonetheless, some more practices such as the so-called gift authorship may still not be detected. Gift authorships are given to influential scientists for purely honorary reasons and clearly artificially increase the weight and centrality of these researchers in the co-authorship network. On the other hand, the opposite of gift authorships are ghost authorships that occur when the contribution of some authors is not recognized. In this way, they remain invisible to co-authorship analysis. Similarly, the authors list may not include all those that actually collaborated to the research process. As a sociological research as shown, not all scientific contributions result into co-authorship of papers. Laboratory technicians, for instance, are rarely included in the authors list. Sometimes even the work of graduate students is not awarded with the authorship. A further phenomenon that complicates the use of co-authorship analysis is the so-called hyper authorship. In fields such as high energy physics, publications resulting from larger scale experiments may list thousands of authors. The paper that reported the discovery of the Higgs boson, for instance, counted more than 5,000 of authors. In social sciences and humanities, we observe in a certain sense the opposite phenomenon because co-authorship is scarcely diffused in these areas. However, the lack of co-authorships does not mean that scholars in social sciences and humanities do not collaborate. In fact, they informally collaborate with many colleagues and the rate of collaboration is similar to that of their colleagues in natural sciences. Still, since these collaborations are not recorded in the authors list of the publications, they remain invisible to co-authorship analysis. With the results that co-authorship networks in the humanities are too sparse to be truly informative as the instance on the right side of the slide shows. As you can see, most of the notes in the network are isolated with only few chains of co-authors appearing in the lower part of the graph. In the light of these shortcomings of co-authorship analysis, scholars in library information sciences have proposed to focus on the acknowledgments of scientific publications to gain a richer picture of collaboration ties in the scientific community. The idea is simple to switch the focus from the byline of the articles, that is the authors list, to the acknowledgement sections of publications. Even first glance at these texts already reveals that they are a very rich source of data about informal collaboration patterns and more generally the social context that surrounds academic articles. As you can see from the example on the screen that is taken from an article in economics, the acknowledgments mention several persons that contributed in various ways to the article, the so-called acknowledges. These mentions allows us not only to reconstruct informal collaboration patterns, but also to find out who are the persons that are most influential in a field by counting how many times they are mentioned in the acknowledgments. However, the acknowledgments shed light also on the institution that provide financial support to the research and lastly, on the conferences and meetings where research is discussed. In some of the acknowledgments are a very interesting source of data for sociological investigation on the social structure of science. Now, in this talk, I will present you a method to analyze some of the information contained in the acknowledgments in order to gain a new insights on the social structure of science. The idea is to model the acknowledgement data as a two-mode network made of articles here on the left and acknowledges here on the right. Then these notes are linked together by the relation of mentioning that these the article notes are connected with the acknowledges notes when the former mentioned the latter. From this two-mode network, we can generate a one-mode network where only the acknowledges notes appear using the method of biographic coupling. That these two acknowledges are connected when they are mentioned together in the same acknowledgments. In this case, for instance, the acknowledges A and B are linked together because they are commensioned in paper one. Similarly, acknowledges A and B and C are connected because they appear together in paper two and acknowledges C and B are linked because they appear together in paper number three. And lastly, we attribute a weight to the links between acknowledges based on the number of papers in which they're mentioned together. So in this case, acknowledges A and B share a link of weight two because they're mentioned together in two papers, namely paper one and paper two. In this way, we have generated the Analogy Pro-Mention Network or ACM network in short. Clearly this network can be partly analyzed using standard tools of network analysis such as centrality measures and arguing for detection of communities. Now, so far I have delineated the general concept of ACM network. However, to better assess the viability and potential interest of this method, I tested it on a humanities area that is analytic philosophy. And in this section of the talk, I will present to you the results of these experiments. Before passing the results, however, let me quickly summarize you the three reasons for choosing analytic philosophy as a case study. First, previous studies have shown that the percentage of articles in analytic philosophy that feature an acknowledgement is quite high. For instance, in 1990, 94% of the articles in mind featured an acknowledgement. Second, the intellectual structure of analytic philosophy has already been mapped with co-citation analysis. Therefore, we have material for comparison between the intellectual and social layers of the field. And lastly, analytic philosophy stands as a good representative of humanities areas more generally. So testing ACM network on it may be useful to assess the viability of the method for other areas in the humanities and maybe social sciences. On this slide, you can see a brief overview of the corpus I used in the analysis. It comprises 2,073 articles published in five of the most prestigious journals of analytic philosophy, a philosophical review, journal of philosophy, mind, noose and philosophy and phenomenological research. The articles span over a period of 15 years from 2005 to 2019. So we have to do with the recent analytical philosophy. To generate the ACM network, I implemented a procedure made of several steps. First, I collected the full text of the 2000s and more articles from which I extracted manually the acknowledgement sections. Note that the place of the acknowledgments in the text is not fixed but changes over time and by journal. Sometimes they are placed in the first note of the article, sometimes in the last, sometimes there is an acknowledgement section and so on. So the problem is that no automatic procedure for extracting the strings of text was possible. I had to manually collect each acknowledgement text which is, as you can imagine, quite a time-consuming operation. However, once each string of text was collected, I automatically extracted the names of the acknowledges from them and this was done by using a named entity recognition module of Spacy that is a Python package for natural language processing and in particular name entity recognition utility is able to recognize named entities from strings and classify them in several categories including persons, organizations, companies and locations. However, the data produced by this system could not be used in the raw form because of two main problems. First, the acknowledgments are frequently written in an informal style and diminutives are used instead of the full form of names. So these names variants must be reduced to standard forms in order to produce reliable results. Hence, a first step in the data cleaning was the merging of variants. Again, this was a quite time-consuming operation. And secondly, sometimes the algorithm for named entity recognition makes errors in the categorization of entities producing both false negatives and false positives that must be corrected manually. After this quite long process of data cleaning, the final dataset was characterized by the following statistics. 94% of the articles in the corpus feature unacknowledgments. The 2000 articles are written by 1,391 distinct authors that mention 5,774 distinct acknowledges. The maximum number of acknowledges mentioned in one paper is 72, quite high number, but 127 articles mention only one analogy. In fact, the average article mentioned 8.3 analogies but there is a great variability in the number of acknowledges per paper as the high standard deviation shows. Interestingly, the average number of authors per paper is significantly lower with only 1.2 authors per paper and only 12% of the papers have more than one author. The statistics confirm that co-authorship is scarcely practiced in analytic philosophy. And let me say as a side note to the presentation that I present and discuss more statistical analysis of this data in an article that should appear in a special issue of logic analysis. So if you're interested in the print, please write me and I will share it with you. Coming back to your presentation, we see in this slide how the low incidence of co-authorship renders co-authorship analysis quite useless for these corpus. As you can see, the network is too sparse for bringing truly informative about the social structure of analytic philosophy as 70% of the authors in the corpus are isolated. The ACM network on the other hand is more informative and interesting to explain. And as you can see it in this slide, in this map that is produced with a software VOS viewer the acknowledges with 10 mentions or more in the corpus are represented with their mutual co-mention relations. In the visualization, the size of the nodes is proportional to the number of mentions an analogies receives, whereas the relative positions of the nodes reflect the co-mention similarities. So that two analogies that are frequently mentioned together in the corpus are placed closer on the map. Lastly, the color of the nodes represents the cluster they're attributed to by the community detection algorithm of VOS viewer. As you can see the ACM network is quite tight with the highly mentioned acknowledges such as David Chalmers and Timothy Williamson placed at the center of the map. Now the community detection algorithm integrates five main clusters which interestingly can be mapped to some sub areas of analytic philosophy. For instance, the red cluster hosts several metaphysicians, the blue clusters philosophers of mine, the green cluster philosophers specialized in ethics and moral philosophy, the violet cluster philosophers of language and several formal epistemologists. Note, however, that the association between philosophical areas and clusters is somehow weakened by the fact that epistemologists are found in all clusters. And this seems to indicate a special role for epistemology in recent analytic philosophy. If we move now to a map of the intellectual structure of analytic philosophy, you can see it on the screen. The second map is generated with co-citation analysis and is based on the same corpus of articles analyzed before. This time, however, the notes represent the references cited in the bibliographies of the articles. The more two references are cited together in the bibliographies, the more they are cut close on the map. Again, the size of the notes corresponds to the citations received by references in the corpus whereas the color is attributed based on the community direction algorithm. Co-citation network like this are useful to reconstruct the structure of the knowledge base of scholarly fields. And in fact, the cluster can be easily mapped to the sub-disciplines of analytic philosophy. The red cluster corresponds to metaphysics, the violet to philosophy of language, the blue to philosophy of mind and the green to moral philosophy. These clusters almost match those we found in the ACM network, but with a notable difference. Epistemology here is represented by a specific cluster. Whereas in the previous network, it was so to say spread all over the clusters. If we compare now the social and intellectual structure that is the network based on the analogies co-manages on the left and the network based on the references, co-citations on the right, we can note how the former is somehow denser than the latter. And this is confirmed at the quantitative level by the statistics of modularity. That is a network statistic that measures how much network is divided into sub-communities. Networks with a high modularity have dense connections between the nodes within clusters, but sparse connection between nodes in different clusters. As you can see on the slide, the ACM network has a lower modularity compared to the co-citation network. Confirming the qualitative impression of higher density we had by inspecting visually the maps. The four based on these results, we can argue that the social structure of analytic philosophy is somehow more interconnected than its intellectual structure. That is that differentiation into philosophical sub-disciplines is more pronounced at the intellectual than at the social level. Another important difference between the two networks is their temporal focus. The ACM network captures the present social structure of analytic philosophy. All the acknowledges that appear in it are in fact active members of the analytic philosophy community. In the sense, the number of mentions they receive reflect their contemporary prestige or symbolic capital in the community to use some sociological concepts. The co-citation network on the other hand focuses on the past as it captures the knowledge base of the discipline. Several of the authors with a high number of citations that appear in this second network, such as David Lewis, for instance, are no more active members of the analytic community because they are that. The fact that different philosophers appear in the two networks tells us that the distribution of prestige does not coincide with the distribution of intellectual capital. That is that the social intellectual layer do not totally coincide. This difference is confirmed if we look at the distribution of mentions and citations shown respectively on the plots on the left and on the right of the site. These plots tell us, to say simply, how much the distribution of a variable in the population represented by the red line deviates from perfect equality, which is represented by the line at 45 degrees. From these plots that are called Lorentz-Turk, we can say, for instance, that the bottom 60% acknowledges receives only 20% of these Lorentz-Turk mentions. Now, as you can see, the distribution of mentions is more distanced from the line of perfect equality than the distribution of citations. A measure of this distance is the genicoefficient, which is a measure derived from this Lorentz curve that quantifies inequality. As you can see, the genicoefficient is a higher for the mentions distribution than for the citations distribution, showing that the symbolic capital is more concentrated in a smaller segment of the acknowledges population than intellectual capital. Okay, so let us now take stock of this quite long journey in the acknowledgments of academic publications. I hope to show you that the acknowledgments analysis and the ACM network are useful methods for investigating the social structure of squaring fields. However, we should point out both its advantages and disadvantages. The first advantage of the method is that it's more effective than co-authorship analysis for tracing the social structure of areas in the social sciences and humanities, as I think it should be clear now. The second advantage is that it allows to consider informal collaboration that remain invisible to co-authorship analysis. It can therefore integrate effectively this type of analysis. However, there are two main issues that make the method difficult to implement in practice. First, there is no acknowledgement index as there are citation indexes like Web of Science or SCOPUS. The extraction of acknowledgement texts from publication is therefore a very time-consuming operation. I must add, however, that in the last years both Web of Science and SCOPUS have started to collect the acknowledgments texts of academic publications. Unfortunately, however, they consider only those acknowledgments where financial support is mentioned. This is done for the funders, basically. However, in the case of humanities, the portion of acknowledgments that mention funding is only a small portion of the acknowledgments. Therefore, for the moment, there is no easy way out from manual data collection. And the second big issue is the huge amount of data cleaning that is needed to consolidate the raw data. As we have seen before, the same person is mentioned with several variants that must be normalized to produce reliable statistics. And the naming and recognition algorithm make mistakes that must be corrected. Also, this operation is very time-consuming and difficult to scale up. Okay, so let me now conclude with some further step in the research that I'd like to perform. So the idea is to basically integrate intellectual and social similarity, starting from a three-mode network like the one you see on the slide. And this network, there are articles that cite references on the left and mention homologies on the right. The idea is that from this network, we can derive two new networks, the first based on bibliographic coupling in which publications are linked together when there is an overlap in their references, in the references they mention. So this is the first network that allows you to reconstruct the intellectual similarity between publication. The other network is the opposite of the analogies convention network and we may call it the analogies coupling network. In it, two publications are linked together when they mention the same analogies and the weight of the link is equal to the number of analogies they share. Now, the idea is that these two networks that rest upon the same sets of nodes can be fused by the new techniques of network fusion that allows you to, you know, match and fuse the edges in the social similarity network and in the intellectual similarity network. However, I still have to perform this analysis. So for the moment, I do not know if the results will be interesting or could be somehow interpreted. However, I hope to show you that these methods based on an advance can be thoroughly integrated with other already known methods in science mapping to have a richer picture of the social intellectual structure of scholarly fields. I have to conclude now. Thank you very much for your attention and please feel free to contact me at these addresses. Thank you. Great. Thank you so much. Very, very cool stuff. Yeah, I'm excited to see where this goes, hopefully. And in all, at the amount of data cleaning that you had to do, that's very impressive. Let me go to the questions now. First, I have a question coming in from Catherine Herfeld who asks, so given that acknowledgments can signal very different kinds of relations and you mentioned this near the end with respect to the scope of state of like funding or feedback or what have you, have you thought about a way to distinguish different kinds of acknowledgments, relations and maybe categorize them or think about their relationships? Yeah, thank you very much. This is a very key question in the analysis of your knowledge. And so from a certain point of view, we have the same issues we have with co-authorship analysis because we have a text that mentioned a bunch of persons and it's difficult to reconstruct a posteriori what were the contributions of each of these persons. Sometimes the authors explicitly say what is the contribution. So for instance, they thank someone for such assistance or for feedback. But sometimes you have only lists of names, basically. I am gratefully acknowledge the help of X, epsilon and Z. And in this case, I think that you just cannot reconstruct what were the contribution. But maybe by performing some advanced text analysis methods we can link somehow the names to some keywords. What I've done in the work I mentioned that should be appeared in a logic analysis was to analyze what were the main, the most frequent keywords that appear in an argument text. And this relates all to gratefulness to thanking people. So it appears that at least in late philosophy the most frequent contribution is a sort of intellectual feedback that people gave. But clearly there is no experiments or no laboratory activity. For instance, in biological sciences, people are thanked also for contributing with samples, for instance, or with statistical analysis. And so for the sciences, it's more rich, the information because it depends on the kind of research practices that they're buying a field. Sure, yeah, that makes a lot of sense, thanks. Another question coming in from Rose Trappes who asks so is there any sociological research that's been done on how people decide who to include in the acknowledgment? So I'm wondering if those acknowledged actually have more social capital in the field or if they just give comments on heaps of drafts. Yeah, so in the field of the sociology of science there are not so many studies focusing on arrangements but in library information sciences there are plenty of them. The main author here is Blaise Cronin that is one of the greatest scholars in library information science and he has basically developed the field of acknowledgments studies. And yeah, so let me say this, in information library sciences, there is a sort of naive approach to acknowledgments so that they are taken at face value. So as a science of informal collaborations. But it's true that for instance, people that has studied acknowledgments in economics thinks that people are mentioned for strategic reasons, for instance to influence the editors in the choice of the reviewers. So there is this idea of cite your friend and mentioning the acknowledgments your foe so that the editor would not choose it as a reviewer. But still there is not so much research on this. I'm trying in another research to find out if there are some quantitative tests that can say to us whether mentions are given because of the social capital or the intellectual contribution of people. And there are not easy tests, but for instance, the great, the skewed distribution of mention that is the fact that few people receive lots of mentions, whereas most of people is only one mention, this is a sign that maybe a cumulative advantage mechanism is working so that prestige has a role in who are the people that are mentioned. And the idea is that sometimes you can signal that you are acquainted with the great scholars in your field to show that you are the right person, that your paper maybe is more important to read or cite. Sure, yeah, that actually connects to, so I'll take chairs prerogative here and ask my own question, even though I didn't get a chance to type it into the box. That relates to a question that I had. So I wonder, yeah, I wonder whether, so there's a number of competing hypotheses, right, that you might envision for what's driving someone being central in a central node in this network. And so like I was thinking for instance of, on the one hand we have sort of people with a lot of social capital or prestige so it's important to impress, but on the other we have something that actually stood out to me from a couple of the names that were central in the network that you showed everyone. Having a lot of visiting positions or running an institute or something, the kinds of things where like, if you're the kind of person who does the kinds of things that you get a lot of people around you all the time that you can tend to bubble up. So I saw in there a couple of names people who I know have big collaborative groups always have a bunch of rotating scholar positions in their community. So I wonder if you've thought about evaluating, I mean, evaluating those hypotheses isn't easy. It's almost a sociological question. I'm just kind of interested to know what your thoughts are because I'm sure you've been thinking about that question. So basically all these issues or questions that are about the process behind the generation of mentions, okay. These are really crucial in the studies of acknowledgements and I think they are very good because in all the quantitative studies we do in digital humanities because sometimes we take the data at face value whereas there are many hypotheses behind their, how they are generated, especially when we study things that are produced by people like mentions and acknowledgements but also words and texts that are used in topic modeling. So the reason why it's very complex social reality behind these products and this is where social epistemology or philosophical scientific practice and social studies of science, they come in. And I think that sometimes the only way to rule out what is the best explanation to find out what is the best explanation is to supplement the quantitative analysis with some qualitative analysis. So asking people, doing interviews, doing some ethnographic research in the communities and clearly there is also expert or task knowledge that these people in the field clearly know some dynamics that from an external point of view are not feasible when you just inspect the network. At the same time I must stress that also these qualitative methods have some issues because for instance, if you interview people sometimes they cannot be very sincere in what are their motivations or they just cannot remember why did they mention that person? So there are always issues. I think that integrating qualitative and quantitative can somehow provide you a richer insight. From the quantitative point of view, what I tried to do was to collect several information, several data about each of them. For instance, the number of citations they get in the web of science or the number of publications or if they are working in English speaking institutions on and they run a multivariate regression analysis model. And from this kind of analysis you can see correlations between variables. And for instance, it appears that in this corpus being affiliated with an English speaking institution give you a sort of bonus in the number of mentions. So it's more likely to receive higher mention if you work in the US or in the Australia or UK. Well, we all know that analytical philosophy is mainly based in the English speaking world. So maybe this is not unexpected but at the same time we have the quantitative, we can quantify measures so to say the effect of the English speaking focus of the community. Sure, yeah, that's really nice, thanks. I don't have another one of the box so I'm gonna help myself again. I wanted to ask about these network fusion methods that you mentioned right near the end. I know that that may be sort of cutting edge stuff for you so I don't wanna put you on the spot if you don't really wanna go into it but I haven't really run across these yet and I wonder what kinds of networks do you expect to get? So how does that process really work exactly? What do you start to see when you think about sort of mixing these networks together? That's something that I've ever really considered. Well, let me say that I don't have a clue about it because I still perform and actually to study that the packages that have been produced to this natural fusion technique that is very recent. I think that the papers in nature about it are coming really in these recent years. So what I know is that my supervisor in Siena actually have worked on the fusion of network of journal networks where the two networks that were refused were the consultation network of the journals and the network based on the editorial overlap. The overlap on the editorial board of journals. Okay, so again, a measure of intellectual similarity and a measure of social similarity. And the results are very good because the two network have a very similar structure. So it seems that the fusion actually provided with this cluster is social intellectual clusters so that the two, the intellectual and social dynamics seem to go in the same direction. So we would like to reproduce these kind of analysis for analytical philosophy and also for economics. I would like to test the results if there are... But for the moment, I must say that I have no idea. Cool, yeah, no, that's a good answer. That's totally fine. It just sounds really neat, I agree with you. I think that's something to be fun to pursue. Question from Chris of Malateru asks, which is a very cool way to derive social networks. Did you think about comparing the ACM networks to networks derived from topic modeling? So he says on our corpus, we conducted a proof of concept for generating social networks on the basis of shared topical distributions in articles. And the same thing could be done on your corpus, I guess, but require having ideally access to the full text or at least to abstract. So I'm not sure what you have in that sense in your corpus. Yeah, so in my particular corpus, I have only the classic metadata. So I have titles and abstracts that I think are not enough for topic modeling. So actually when I extracted the acknowledgement text, I had to access the full text, but I could not download all these texts for bigger reasons. And so this is for the practical side. From the methodological theoretical point of view, that would be definitely very interesting because this idea of integrating, so again, I always speak in terms of layers, but we know it's just an analytical image. There's various dimensions of the research enterprise that is the level of ideas, so to say the level of topics with a level of the persons that embodies, so to say this idea. I think this is very interesting. And also for any history of science that they are doing these, there are very interesting work on, for instance, the history of general relativity, where they try to mix all these networks based on citations, based on working on same institution, participating in the same conferences. People at the Max Planck Institute in Berlin are doing these. So I think that many enterprises, similar enterprises are performed now in very different areas. And I think one of the most important things to do is to talk to each other, like this conference is doing. So, well, yeah, I'd really like to do that. At the moment it's not in the plan, but in the future, yes. I actually was slated to go do a research stay at the Max Planck in Berlin. It was canceled to COVID. So yes, I think talking to each other, I know the feeling, hopefully it will be able to have more opportunities to do that as things start to get back to normal. Thanks, let me see just real quick, if there's anything else that's come into the Q&A box. You would have time for maybe one more quick question if somebody wants to throw something in. I'll just stall for a few seconds to let the tape delay catch up. This is all fancy stalling tactics. At some point the audience is going to start realizing when I'm filling time. But no, this is just a moment. Well, if there's not any further questions, we're very close to time now. So let me go ahead and stop it there. Thanks so much, it's very cool talk and really appreciate it. We will be back in with a 10 minute break now. So a 10 minute pause and we'll be back with the next talk shortly. Thanks very much.