Category Archives: Algorithmic Criticism & Emblems

Modelling, McCarty and Me

Computing has no significance whatsoever for research in the humanities other than in what is to be done with it, concretely and immediately. … innumerable directions may be taken. … which do we take? (McCarty 199)

Though in the above quote McCarty questions what the set of agendas for the field of humanities computing should be, he also argues individual scholars engaging with particular works are a “primary” driving force in the field (206). Given that, I feel free to ask myself what my agenda is, what direction I desire to take (see McCarty 200 on desire and planned action), as I examine the relationships between the symbols in Renaissance emblem books.

McCarty diagrams the stages of modelling from cultural artifact [emblem books] into computer science and back up to the artifact through humanities computing (197). It seems I have already begun the first two movements: problem specification, and systems analysis. With the first step I have already decided my “specific perspective”: I wish to look at the symbols related to and with the Judeo-Christian God. In a vague way, I already conceived emblem books as “a system of discrete components [the symbols] and relations [within and between emblems]” (McCarty 197). McCarty correctly states “sacrifices” and “compromises” are made during these steps (198). By picking one perspective, I limit what I am studying about the work. How I choose to study the system of symbols affects what sorts of results can be gained, at least from this project.

Yet, because of McCarty, I no longer think my “conceptual design” is “good enough” (see McCarty 198). Originally, I wanted to use the Iconclass hierarchy metadata (see McCarty 89-90 on metadata), to group symbols together in specific emblems and to see which groups persisted in many emblems. In general, projects using Iconclass tend to be very thorough regarding the “what.” For example, looking at this emblem “Invidia [Envy]” as catalogued by the Alciato at Glasgow project, one can see they have missed very little (if anything) with their Iconclass tagging. Reading McCarty’s description of his project looking at personification in the Metamorphosis,[1] lead me to questions of context (54) and weighting (61).[2] Contextually, I want to know what part of the emblem – motto, pictura, and/or subscription – the tags are pointing towards. Within the image itself, the element placement itself could have some contextual meaning: Is the main focal point of the image frequently in the centre? Perhaps I could experiment with image analysis/design software such as Pixcavator IA or PhiMatrix to generate data (taking the “off-the-shelf” route [McCarty 197]). Weighting wise: Should a connection between Iconclass tags be weighted more based on type of relation; in the example, should tags describing envy have a stronger link between them than with the “scenery”?

Like McCarty’s “blind man” I can use received knowledge, in the form of tools (Iconclass and software) and instruction (classes and reading), but my DH work is still self-moving and experimental (McCarty 51). I will learn by doing, and adjust direction based on interaction with the programs and data, making sure to remain explicit about and consistent in applying my choices (McCarty 25).


Works Cited

Adams, Alison et al. Alciato at Glasgow. University of Glasgow. Web. 9 Dec. 2015.

McCarty, Willard. Humanities Computing. Basingstoke, Hampshire: Palgrave Macmillan, 2014. Print.


[1] Oddly, my emblem example is a personification whose subscription comes from Ovid’s Metamorphosis. That was not planned.

[2] McCarty’s application of Heidegger to the process of modelling is quite apt here. For me, the Iconclass tool has gone from “ready-to-hand” as Heidegger defines a tool in use, to “present-at-hand” due to “a failure of the tool” to do what I want it to (McCarty 42), though admittedly not what it was designed for. I have to modify the tool. I expect this will happen more than once, as has been the case in my previous and current DH projects.

Graphs, Maps, Trees: Thoughts on Moretti

            Unlike McCarty, who tended towards generalization, Franco Moretti uses specific projects as exemplars, showing the possibilities of quantitative data and distant reading for the humanities. While my focus is on the “Graphs” chapter, the other two sections impacted my thinking as well. “Trees” made me consider diagramming the representational differences occurring for a single symbol: Dog – Ears Up/Down – Sitting/Standing etc. (see Moretti 77). In “Maps”, Moretti argues that the locations on the map were not at “significant” as the “relations” revealed by the diagrams (54-55). My project is akin to mapping because I wish to look at the relations between symbols; and I could use weighting (see last post) to indicate physical/metaphorical closeness/distance; for example, when the Iconclass tags all refer to aspects of the same ‘figure’ or when a literal image stands in for a metaphorical idea.

            While my project is quite different from Moretti’s study of the rise and fall of genres, several of his comments in “Graphs” are applicable. In general, his comments on the disappearance of genres (18) made me wonder: “How ‘stable’ are the emblem symbols over time?” More importantly, Moretti states “[Quantitative research] provides data, not interpretation” (9, emphasis in original).[1] The data is thus another text, a meta-text, which, akin to close reading, is the object of interpretation. Moretti also mentions how he gathered his data on genres, using the multiple sources and comparing them (18). Unfortunately, while I will be gathering data from other scholars, the Iconclass tagging projects tend not to overlap. Thus, I am limited to one set of tags for most books. Perhaps this fact is mediated by internal double-checking by different scholars within the projects, but it is still something I have to deal with when gathering data.

            Additionally, Moretti discusses the difference between “individual events” and “patterns” each requiring different explanation types (13). I suspect both of these types will occur in my project: symbols that are unique to one author/illustrator, and ones that appear and group repeatedly. Finally, Moretti admits to an instance of not being able to explain something presented by his data (26). This led me to the profound yet simple thought: “It is okay if I do not have an answer.” Such a situation is grounds for further questioning and research.


Works Cited

McCarty, Willard. Humanities Computing. Basingstoke, Hampshire: Palgrave Macmillan, 2014. Print.

Moretti, Franco. Graphs, Maps, Trees: Abstract Models for a Literary History. London: Verso, 2007. Print.


[1] Moretti states that “ideally” data should be “independent of interpretations” (9). That “ideally” hints at, but does not enter into, the problem of interpretation during the data gathering phase. Here is where McCarty’s emphasis “complete explicitness” regarding what one is doing (choice-wise for example) and “absolute consistency” come into play (McCarty 25), hopefully bringing the data as close as possible to the ideal.

A Problem of Standardisation

            My project has a problem. It is one I discovered while researching the Iconclass tags for the Christian deity. Specifically, the issue is whether or not the “single mark-up scheme” – in this case Iconclass – has been applied “thoroughly and consistently” (Cohen and Rosenzweig). In my first post, I praised the thoroughness exhibited in the tagging of an emblem; now I offer one where that level of thoroughness is lacking. This emblem is complex,[1] and at least one element I am interested in has not been tagged in the “Iconclass Headings” section: the tetragram (Hebrew letters representing God the Father – Iconclass tag 11C13) found top-centre near the Holy-Spirit-as-dove. Additionally, there can be multiple tags for similar things: The tetragram could also be tagged 12A111 indicating its use in Judaism.[2] The crucified Christ in the upper-left corner (labeled C in the emblem itself) could be tagged under symbols of Christ (11D) or in images of the Passion (73D), as done here.[3] So, I am relying on the choices and mistakes of others if I use their tags; and I am concerned that this could compromise my results.

            However, there may be no other (expedient) way.[4] Both John Lavagnino and Matthew Jockers offered insight into this problem as I was thinking about it; the latter quite indirectly. Lavagnino writes about how text is different from the elements of a painting (“The Nature of Texts”).[5] He argues you can break text down into “discrete parts”: words or letters. Then you can search through a text, for example, by counting the use of those parts. BUT: “There is no easy way to decompose digital images into anything like an alphabet” (“The Nature of Texts”).[6] Iconclass deals with this ‘searchability’ issue to a certain extent, though the tags are more akin to abstract notions and ideas rather than words or sentences, because Iconclass allows one to tag an image with its associated meaning (see the section on “keys” here, for example). Still, given that one cannot search parts of images directly, that makes the issue of thorough and consistent application paramount. Jockers, quite indirectly, indicates my problem.[7] Writing about TEI, Jockers states “the amount of metadata available [using the TEI markup scheme] is only limited by the encoder’s willingness to modify the documents” (section 10.2). Iconclass is different, but that statement holds true for it as well. Different projects (and different individuals within a project) can have different standards. One project could just tag full human figures for example, while another may get into body parts, positions and expressions. Since I am using metadata from many projects, I cannot guarantee that the Iconclass tags have been applied with the same standard of thoroughness and consistency. Therein lies the problem.


Works Cited

Cohen, Daniel J., and Roy Rosenzweig. “Becoming Digital: To Mark Up, or Not To Mark Up.Digital History: A Guide to Gathering, Preserving, and Presenting the Past on the Web. Center for History and New Media, n.d. Web. 2 Jan. 2016.

Contents of Iconclass. Iconclass. RKD, 2012. Web. 13 Jan. 2016.

Jockers, Matthew L. Text Analysis with R for Students of Literature. Cham, Switzerland: Springer, 2014. Print.

Lavagnino, John. “Digital and Analog Texts.A Companion to Digital Literary Studies. Ed. Ray Siemens and Susan Schreibman. Oxford: Blackwell, 2008. Web. 12 Nov. 2015.

Stäcker, Thomas et al. Emblematica Online: Resources for Emblem Studies. University of Illinois and Herzog August Bibliothek. Web. 13 Jan. 2016.



[1] Click on “View scanned image of Emblem” to open the image in another tab.

[2] However, such a tag may be inappropriate here given the Christian context.

[3] I might add that I believe the latter was the correct choice, given the specific options.

[4] I could tag all the emblems myself, if I had the time. Of course, there would likely still be errors, as I could very well miss elements or incorrectly tag things.

[5] This is a section heading of the online text.

[6] I should note that Lavagnino uses this observation to argue that unsearchable elements, such as font selection and page spacing – that is elements of form – also potentially have importance for the “aesthetic dimension” of the text.

[7] I wish to note that I quite enjoyed Jockers’ book “Text Analysis with R for the Students of Literature” as it excited me with the possibilities for using DH tools. However, most of my responses to Jockers were based on analogy/simile from what he was discussing to what I wanted to do. For example, Jockers wrote about the coding needed to show that multiple text files held works written by the same author, thus creating the “vector of author names” (section 12.7). In the margins of the section, I wrote “I need a way of showing that Iconclass subcategories [analogous to individual text files] are types of the primary categories [analogous to the vector of author names]; but unlike this instance I want both the larger category and the smaller specific type.” I include this to show that my mindset while reading Jockers’ text was not that of following him to the ‘t’; rather I let my mind wander using his text as a springboard to problems and ideas for solutions. Hence the tenuous connection I’ve included in the main blog post.

The Challenge of Hoover and McGann

[Note: When I originally wrote this post it was simply my reaction to David L. Hoover’s essay “Quantitative Analysis and Literary Studies.” The ending of the post was rather weak, largely because I was discouraged by my own desire for definitive and objective data within my project and the realisation that, at present, it does not exist. Perhaps I had fallen into the cognitive distortion of “All or Nothing” thinking,[1] where because it could not be perfect, I viewed my project as a total failure, unable to provide any knowledge about the emblems. So, I delayed posting it.

During the delay, I began reading Jerome McGann’s book Radiant Textuality: Literature after the World Wide Web. Hope returned, bringing with it a more realistic sense of how things stand. What follows is partially the original post, now edited with some references to McGann. As a nod toward McGann’s work with online archives, I have decided to clearly indicate my edits: additions are italicised in square brackets; deleted sections have been retained in footnotes. My goal with this experiment, is to retain both interpretations, to explicitly show my subjective view, simply because I believe my project, and other quantitative humanities projects, benefit from such candor, as it makes interpretive decisions more apparent for the end-user/reader. Thus, it seems a good idea to begin the practice now, showing what was changed much like the “historical log” of project changes McGann discusses (92-93).]

Given the problem I discussed last post, it is of little surprise that reading David L. Hoover frustrated me further. Hoover differentiates what close reading and quantifiable analysis can do, and thus what sort of arguments one can make based on the approach. Additionally, he emphasises the “numerical” nature of quantitative approaches especially the element of “accuracy” inherent in mathematics in “measurement, classification, and analysis” (“History, Goals,…”).[2] One thus gets the idea that Hoover wants to separate an individual’s interpretations from the gathering of the data itself. Such a reading is bolstered by his emphasis on automating steps (“Methods”) and his critical response to a quote regarding using “traditional critical interpretation” to “understand the meaning of those statistics.” Hoover indicates that such an approach could be misleading, though he does not seem to discount it entirely (“Four Exemplary Studies”).[3] [Hoover wants a high level, or even complete, objectivity; the objectivity of mathematics (“History, Goals,…”). Jerome McGann counters that “Objectively considered, objectivity is impossible” and the pursuit of it potentially a hindrance (24). What can/should be aimed for is “self-conscious subjectivity” (24), that is, a high-level awareness of one’s interpretations, influences and choices. Such a goal includes, ideally, an accounting of those choices/standards to one’s end-user/audience, allowing them to make a more informed decision as to the usefulness of your data and interpretations. That is not an easy task; Emblematica Online, for example, discusses that it will use Iconclass on its “About the Project” (last paragraph of “Metadata” section). However, I haven’t been able to locate a statement regarding how thorough and complete they intend to tag, or what decisions/guidelines they mandated in their tagging.[4] Where Hoover would likely discard the data I intend to use, McGann would likely accept it, granting that there are challenges to over-come as I go.]

Hoover presents several challenges in his first two sections. In discussing Woolf, he states that to claim something as “characteristic” as “unusual” requires quantified comparison (“History, Goals…”). I would like to know which symbols/symbol complexes are part of larger patterns and which are restricted to a particular work; but I cannot, at present, compare all 1,388 emblem books on Emblematica Online simply because not all of the emblems have been tagged, nor is there a program that can accurately search images for specific parts (given differences in drawing, for example). So perhaps I cannot make claims regarding what is truly a pattern for the whole genre, but I could make a statement regarding the texts I do sample, presenting it similarly to how Hoover does “in a corpus of 46 Victorian novels by six authors…” (“Methods”). Hoover is right when he says that anything “that can be reliably identified can be counted” (“Methods” my emphasis). The issue is inherent in the word “reliably”. Even Hoover admits that “semantic categories”, which images as signs/symbols arguably fall into, are “more difficult to count” (“Methods”), likely for that very reason. [While Hoover proposes an automated system to deal with “semantic categories”(“Methods”), McGann discusses the need for tools for individuals to mark up images, specifically allowing parts of images to be accessible to researchers in a searchable, and likely countable, way (62, 69, 94-95). McGann, unlike Hoover, seems to favour a hands-on approach in this regard. McGann leaves room for the human.]

[5][Originally, I bought in to Hoover’s statement that “poor initial choices can lead to wasted effort and worthless results” (“Methods”) and I considered whether using the Iconclass decisions of others was such a mistake. McGann himself writes “it does not bode well to begin with a logic one knows to be inadequate” (90). It is quite easy to see issues with the Iconclass data I intend to use; for example, individual taggers may have tagged things slightly differently.[6] If I aspire to Hoover’s level of objective certainty, then the issues with this data become unbearable. BUT, McGann’s next sentence applies: “On the other hand, what were the choices?” (90). Additionally, like McGann, I take issue with Hoover’s apparent assertion that failure is “worthless” (“Methods”). McGann writes about the “unexpected rewards of failure” (82). He indicates a growth in knowledge not just of the object(s) studied, but also of the one’s own methods.[7] He even relates an instance where playing, or experimenting, with tools revealed unexpected things, prompting further exploration (82-87). While Hoover seems to want each project to know exactly where it will go,[8]McGann is more open to serendipity, while still arguing for documentation of choices and standards (91-93 for example) and for consistent application (for example, image alterations that appeared useful in one instance, were recorded and exactly applied to others for comparative purposes; see 85). From such a perspective, one cannot know if a previous choice is “poor”, until one works with it, and since one still learns, the effort and results can never be completely “worthless”. I choose to follow McGann, and learn where I am going, as I am going there.]


Works Cited


Burns, David D. “Checklist of Cognitive Distortions.” 1999. PDF file.

Hoover, David L. “Quantitative Analysis and Literary Studies.” A Companion to Digital Literary Studies. Ed. Ray Siemens and Susan Schreibman. Oxford: Blackwell, 2008. Web. 4 Jan. 2016.

Lavagnino, John. “Digital and Analog Texts.A Companion to Digital Literary Studies. Ed. Ray Siemens and Susan Schreibman. Oxford: Blackwell, 2008. Web. 12 Nov. 2015.

McGann, Jerome. Radiant Textuality: Literature after the World Wide Web. New York: Palgrave, 2001. Print.

Stäcker, Thomas et al. “About the Project.” Emblematica Online: Resources for Emblem Studies. University of Illinois and Herzog August Bibliothek. Web. 13 Jan. 2016.



[1] See this PDF for more info on cognitive distortions and “All or Nothing” thinking – particularly the first entry in section “The Ten Forms of Twisted Thinking”.

[2] Citations for Hoover indicate section headings due to the lack of page numbers.

[3] Original post: Since the Iconclass metadata relies on humans inputting the data, it seems that human interpretation of the images is inevitable, even if I get a computer to count the tags I am arguably not counting elements of the emblems themselves, (remember that Lavagnino discussed how difficult it is to program a way to count images, given how an object can be drawn many different ways), I am counting others interpretations, even if I can reasonably argue for a level of accuracy in those interpretations. While I am still questioning if that may offer something useful and meaningful, I don’t think Hoover would approve.

[4] [Even with that, McGann notes problems of ambiguity and interpretation of the markup systems within his own project (91). It seems that there is always some wiggle room, no matter how self-conscious or explicit one makes one’s subjectivity. My preference would be for each tag to ‘highlight’ or somehow indicate which portion of the image it is referring to, thereby giving me, and other end-users, an opportunity to evaluate the choice of tag. Alas.]

[5] Original concluding paragraph: Thus I am left with the question, is using the Iconclass tags a “poor initial choice” which “can lead to wasted effort and worthless results” (“Methods”)? I suppose it depends on what questions I want to answer and what claims I hope to make. Still, perhaps I can counter Hoover by asking, what makes a result worthless, even if a result cannot answer questions because of issues in its method, does it not at least teach us about the holes in the method itself and what needs to be improved? If so, is the effort actually wasted?

[6] [I haven’t been able to confirm or deny whether the emblems were tagged by more than one person each; I hope they were, as it at least gives the opportunity for one person to see and tag things the other missed.]

[7] [Or even the limitations of one’s software (94).]

[8] [To be fair, Hoover does leave room for “happy surprises”, though he seems to indicate that those should lead to a separate study in that they call for “further or different quantification” (“Methods”), rather than an integration into the present project.]

Counting Symbols – A Proposal

Semiotics and religion seem inseparable. Christianity itself has a rich tradition of symbols used to indicate aspects of their deity. Emblem books, a genre which began and flourished during the Renaissance, partook in that symbolic history, using earlier symbols and perhaps creating their own. However, each symbol did not exist in a vacuum on the emblem book page, they united with other characters, objects and settings in a complex of visual semiotics.

Though the Iconclass data of Emblematica Online’s large collection of emblem books is limited, simply because not every emblem has been tagged, I wish to test the available data’s ability to answer the following two questions. First, which particular symbolic representations for the Christian deity (this includes abstract notions of god, the trinity, and each of the three persons) were culturally prevalent (and to what extent), and which were likely quirks of a particular emblem creator? Second, which other tagged image-signs appear most frequently with the symbols for God (whether they have signifiers pointing to physical things or abstract notions) and what can be inferred from their common association? For example, it could be an artistic tendency by emblem creators (for a type of setting, perhaps), a known and still used semiotic linkage between the ideas expressed, or one contained to the cultural circumstances of the genre. While working towards answering these questions, I will also be able to examine if Iconclass tags are an accurate enough data set for quantitatively based literature investigations.

I will compile (hopefully with the assistance of Emblematica Online staff) a data set for a selected corpus of 1686 emblems,[1] indicating the book, author and illustrator (if the latter is available), nation of publication (and language), and year published for each emblem, using this information to map out the currently known extent of use for a particular set of symbols as labeled with Iconclass. For the same set of Iconclass categories, I will compile data on which other Iconclass categories appear with each most frequently and offer an interpretation of why, perhaps in conjunction with a close reading, as needed. I have selected about 50 Iconclass categories that tend towards abstractions and personifications rather than the illustration of biblical narrative, as these tend to be more inclusive of an overall theological picture, rather than just a repetition of biblical scenes.[2] These categories cover abstract notions of God, including the deity’s wrath and role as creator; symbolic representations of the trinity as a group, and for each of the persons: Father, Son and Holy Ghost.[3] These symbols can be geometric – circles and triangles – objects, animals, words and letters, mythological creatures or personifications. Thus they can be studied not just individually, but perhaps grouped as well.

Ferdinand de Saussure explained signs (both signifiers and signified) have “no natural connection with the signified” (854). While a drawing of a tree can problematize this notion,[4] it remains true for symbols of a religious nature. Looking at the symbols for the Christian deity within emblem books, examining how widespread they were in place and time and what pictorial/relational contexts they were used in, can offer a clearer picture of the social conventions regarding religion and spirituality within which the emblem book creators and their audiences lived. Not only may it reveal instances of symbolic inheritance from earlier visual arts, it could indicate potential avenues for symbol groupings that exist even today.



Works Cited

Stäcker, Thomas et al. “About the Project.” Emblematica Online: Resources for Emblem Studies. University of Illinois and Herzog August Bibliothek. Web. 13 Jan. 2016.

Saussure, Ferdinand de. “From Course in General Linguistics.” Trans. Wade Baskin. The Norton Anthology of Theory and Criticism. 2nd Ed. Eds. Vincent B. Leitch et al. New York: W. W. Norton, 2010. 850-66. Print.



[1] This number may be smaller in the end, given that some emblems could include more than one of the symbols I wish to study.

[2] Biblical scenes may still be included, simply because the symbols are used in them as well. In fact, I may discover that, in the emblem books, certain symbols are most frequently associated with such scenes.

[3] AKA Holy Spirit, depending on one’s time period and tradition.

[4] As it could intentionally be symbolising the visual aspect of an actual tree, and thus is not as arbitrary.

Learning from Emblematica Online

            When I placed “Descriptive Metadata, Iconclass, and Digitized Emblem Literature” on my reading list for this project I had no idea, or I had forgotten, that it was written by librarians directly involved with Emblematica Online. Basically, the article discusses issues that they ran into while making the site, and digitising and marking-up the emblems. It also explains the choices they made, such as why they picked Iconclass (it allows for multi-lingual access to the tags [112], and it has a controlled vocabulary [113]). In short their goal was “increased knowledge” about emblem books, and as such the emblems had to be marked-up individually, rather than just digitised. Against the issues I had with David L. Hoover (see this post), the Emblematica Online team recognised that given the “heterogeneous” nature of emblems, determining where an emblem began and ended, as well as “labeling its components requires analysis and interpretation” (114). Humans, specifically librarians and emblem scholars, collaborated on the emblem-level metadata (114). Given that assurance of expertise and reasoning, I am much more confident in the quality of the metadata I wish to use, both Iconclass and bibliographical.

            Cole, Han and Vannoy’s introduction to this paper was quite helpful for my understanding of emblem books; it widened my period of study from just the Renaissance into the Baroque period, covering the years 1531-c. 1750 (111). The trio also discuss the language emblem books were written in, giving credence to my suspicion that the language metadata could indicate where a book was published; in fact, they assure, the vernacular was more common than Latin (111). They also list possible broad sources that inspired the emblem creators: fables, mythology, the Bible, etc. (11). Not that I was unaware of that fact, but it is nice to have a citable source.

            The paper introduced a new term to me: “granularity.” Granularity refers to the “scale or level of detail present in a set of data” (Google definition – the OED failed me on this one; their definition has not been updated, apparently). Basically, the Emblematica Online team dealt with emblem books at different levels of detail: book-level (providing bibliographic details of a full book), emblem-level (112), and even pictura-, inscriptio-, and subscriptio-level (the parts of the emblema triplex [113]). My concern, as I have discovered while using the site, is the transferable access of metadata through different granular levels. Unfortunately, if one uses Iconclass notation as one’s search term, one will only get a listing of applicable emblems; the site does not indicate how many books have emblems meeting that criteria. That means, from a user standpoint, a lot of clicking, counting and copy-pasting; especially since the bibliographic information is locked into the book-level, and does not repeat on the pages at the emblem-level.

            It seems, perhaps, that the type of quantitative project I am undertaking was not originally considered by the team behind Emblimatica Online. Fair enough! It would be impossible to predict every user’s wishes ahead of time. Still, I am encouraged that the librarian trio ended their paper noting that “Collaboration [with Emblem Scholars] is Key” (119). Though this last section seems to say that the scholar is there to ‘fill up’ the metadata information, it also indicates that emblem scholars have helped delineate how “resources” were “sub-divided, identified and made accessible” (119-20). Perhaps I am offering a new problem for the librarians, scholars, and IT personnel to discuss and work to solve.


Works Cited

Cole, Timothy W., Myung-Ja K. Han, and Jordan A. Vannoy. “Descriptive Metadata, Iconclass, and Digitized Emblem Literature.” Proceedings of the 12th ACM/IEEE-CS joint conference on Digital Libraries. New York: Association for Computing Machinery, 2012. 111-20. ACM Digital Library. Web. 28 Jun. 2015.

“granular, adj.” OED Online. Oxford University Press, December 2015. Web. 17 February 2016.

“granularity, n.” Google. Web. 17 February 2016.

Cautiously Reading Data

I recently re-read “How to Read a Literary Visualisation: Network Effects in the Lake School of Romantic Poetry” by Laura Mandell et al. in tandem with Matthew L. Jockers’ text Macroanalysis: Digital Methods and Literary History, which I had not read before. My interest in both stemmed from my growing intention to map Iconclass notations surrounding specific symbols into a network diagram. Jockers offered two helpful tidbits on that front. First, pointing out Gephi, a program I may be able to use (163). Second, and more importantly, he clearly labels what the nodes and edges of his network diagram represent (163), a practice I certainly wish to follow. While neither Mandell et al. nor Jockers helped me understand how I can make my own network diagrams, they both provided some insight into a responsibly cautious approach to reading the visualisations.

Jockers shows his caution in a few different places, but in general it comes down to ‘not making the data say more than it actually says.’ For example, Jockers’ data shows that, within his corpus of 3,346 books, Jane Austen’s Sense and Sensibility is the most similar text to her Pride and Prejudice. He makes it quite clear that there could be another text outside his corpus that is ‘more similar’, and that as works are added to the corpus the data may change (161). Considering my emblem data comes from just a subset of the emblems digitised on Emblematica Online, because only some have been tagged with Iconclass notations, I will have to be somewhat cautious in my interpretations, recognising that what this set of data tells me, could be proven false as more emblems are tagged. In a separate context, discussing word clusters and what they could indicate about nationality, Jockers writes, “The temptation to “read” even more into these patterns is great”, concluding the paragraph with a sobering statement regarding the limits of what the data indicates (115). I believe the caution here is not to confuse one’s interpretation of the data with what the data actually contains. Not that Jockers is against interpretation – “there will always be a movement from facts to the interpretation of facts” (30) – just that he seems to advocate for a clear delineation between the two. I think it is responsible to state what the emblem data shows, and what I interpret that to mean, while not conflating the two or jumping too far from what the data can support.

To that end Mandell et al. warns against mistakes in reading visualisations, explaining that “a major principle of Information Visualisation is that the first thing we will see … is “errors” in our data” (section 1). ‘Errors’ here meaning “information about how the data is structured (section 1). Mandell et al. explain that the data in a visualisation can show things that are not directly related to the research question: the sudden increase of the word “presumption” in an ngram visualisation does not actually reflect the word’s usage, because it does not account for the shift from long- to short-s in typography (section 1). Later, Mandell et al. mention how program bugs miscategorised data, and how a publishing house fire (mid-19th century) removed a swath of data before the researchers even had the chance to visualise it (section 1). For my project, this means I have to grant that my visualisations likely will contain ““artifacts in the data”, and information “about the way it is collected”” (section 1) along with the information I wish to study. I think it would be irresponsible to refuse the possibility that the emblem books tagged may weigh things towards certain authors (Andrea Alciato, the ‘father’ of emblem books, being the prime example) and regions, simply because they are the ones tagged so far. It seems a little caution, both while reading the visualisations and while relating the data and my interpretations to others, is justified. Tempering my enthusiasm with some care likely will help ground my arguments, hopefully making them stronger.


Works Cited

Jockers, Matthew L. Macroanalysis: Digital Methods and Literary History. Urbana, IL: U of Illinois P, 2013. Print.

Mandell, Laura, et al. “How to Read a Literary Visualisation: Network Effects in the Lake School of Romantic Poetry.” Digital Studies / Le Champ Numérique 3.2 (2012): n. pag. Web. 11 Feb. 2015.