Sometimes things do not go as planned. Originally I intended to use a matrix of the Iconclass notations – where the list of notations would be the x and y axis, and the numbers within the matrix would indicate the rate of co-occurrence – within Tableau, and then transfer the data to Gephi in order to examine the grouping of signifiers around each separate God-signifier. Tableau was non-cooperative. However, John Brosz thankfully found a workable solution for looking at clusters of co-occurrences within Tableau.[1] What follows is almost a step by step illustration of the process, along with some of the results. The purpose of this write-up is less about interpreting the results themselves, and more about testing the method. It does show that one can use Iconclass notation co-occurrences to show what emblem elements were grouped together frequently, with generally accurate results. However, it is not problem free, requiring some improvements to the methodogy and the organisation of data.
God’s Wrath – A Test Case
I decided on using the Iconclass notation “11A3 God’s wrath” as my first test case simply because it is quite abstract. I believed the co-occurring notations would thus illuminate:
- How God’s wrath was depicted. My guess: “26E2 lightning, flash of lightning, thunderbolt.”
- The target of His wrath. My guess: “11N31 Pride; ‘Superbia’ (Ripa) ~ personification of one of the Seven Deadly Sins.”
- How that target was illustrated. My guess: “41A3611spire (of tower on house or building)” or “47G91 collapsing building.”
I was right for the first hypothesis in some cases, but not for the latter two. Additionally, I was surprised that the results tend to highlight the direct adaptation and copying of emblems from one book to another. In my desire to discover a general picture of the iconographic language surrounding God’s wrath, I completely forgot about direct intertextuality and the difficulties it would present, despite knowing that it was a practice emblem-makers regularly engaged in.
Set-up: For each emblem in the data set, Timothy Cole and Janina Sarol provided me with the list of Iconclass notations associated with the emblem. John Brosz and I sat down together and created a co-occurrence visualization using parameters and sets.[2] We then duplicated the parameters and sets, so that I could manually adjust four visualisations to map co-occurrence groups of more than two notations. Additionally, we used the separate variable “Requested Notation” to restrict the data set to just show the emblems that include a notation starting with the notation I requested. As noted in Part 2, the data set for the each requested notation includes emblems tagged with any and all of its subdivisions within the hierarchy; which caused one additional manual check that I almost forgot about (as discussed below). The process was workable, but somewhat time-consuming while in use, and occasionally confusing. Still, it offered a window into God’s Wrath.
Step #1: Create a list of the most frequent notations occurring with Iconclass notations that start with 11A3; thereby including “God’s Wrath” and its subcategories.[3] The following graph shows the list of notations that appear in two or more emblems within the limited dataset of emblems including notations starting with 11A3 (Total: forty-eight emblems).
The first column labeled “Null” shows the forty-two instances of the exact 11A3 notation within the set, and also illustrates the first difficulty with this method, because I was looking for 48 emblems total. The problem is that the parameter John Brosz and I labelled “Notation Contains” would only let me look at the co-occurrences of one notation at a time. Therefore, I also have to initially look at the subcategories and their co-occurrences separately, just to manually add their data sets into the list of the most frequent notations. Later on, this ceases to be a problem when I begin to look at the symbols in that list, because the “Requested Notation” filter ensures that all the emblems in the set belong to 11A3 or one of its subcategories.
In the case of 11A3 there are three subcategoreis: “11A3:57AA741(+4)”, which combines God’s wrath with the idea of “revenge”; “11A3(+1)”, which adds the Trinity through the “+1” key; and the subdivision of “11A31 ‘Flagello di Dio’ (Ripa)” which translates to “The Scourge of God”.
The above image shows the frequency of notations for 11A3 and its 3 subcategories. Basically I manually checked each column, adding instances of the same Iconclass notation from the subcategories to those that co-occur with 11A3. However, I found that none of the notations in the three categories contributed enough to change the list of “Most Frequent Notations.” Still, the data is still somewhat interesting. The three emblems tagged “11A3:57AA741(+4)”, top right in Figure 2, share all of the same notations; they have been tagged exactly the same, thus are likely very similar, if not identical. This is a pattern I found again later, which seems to indicate when an emblem pictura is adapted, or directly copied, from another.
The List: Organised from most to least, the following list shows the Iconclass notations that appeared at least five times alongside 11A3 and its three sub-notations. The count of frequency appears in the square brackets at the end of each row.
Step #2: Using the four tableau worksheets, look at which notations frequently appear with each other, within the 11A3 dataset; building each successive graph upon the previous. For example, In the below image, “1st Co-occurrence” looks at the notations co-occurring with 25I1 (limited to two or more co-occurrences). Since the “Requested Notation” filter is set to 11A3, this means all eight emblems are tagged both 11A3 and 25I1, a fact confirmed by the first two columns.[4]
The visualisation shows that the Iconclass notation “25G3 trees” is the most frequent one to occur with both “11A3” and “25I1”. So God’s wrath appears three times[5] with a general city-view and some trees. Not a particularly enlightening fact. Plus, now the numbers are so low, that one could simply look at the three emblems to determine similarities and differences. The 2nd Co-occurrence graph, below, shows the notation frequency for those three emblems, simply by making “25G3” the focus.
Step #3: Repeat step #2 for the second notation on the “most frequent notations list”. I could certainly do the same with the third, forth and fifth notations. However, what I found when looking at “31A2741 throwing something” made me stop and try something else to investigate what I was seeing. This is what I saw:
“1st Co-occurrence”, top-right, shows the notations that most frequently co-occur with 31A2741, within the dataset for God’s wrath (because of the “Requested Notation” filter). “2nd Co-occurrence”, top-right, does the same with 11D51, because it was the third column in “1st Co-occurrence”. Similarly, “3rd Co-occurrence”, bottom-left, shows what co-occurs with 11Q0, and “4th Co-occurrence”, bottom-right, shows the same for 23C311 – because they were the next two columns in the first graph. The four visualisations look oddly similar, showing the same nineteen notations co-occurring four times with my pair.[6] Therefore, it appeared I had a regular grouping of God’s wrath plus twenty other Iconclass notations associated with four specific emblems.
Step #4: To confirm my suspicions, I made a separate worksheet, looking at what I call Emblem DNA, using the unique emblem ID codes to show which emblems had these twenty notations. First, I had to find those emblem ID codes from the notations, using a filter to show just those notations.
The above graph shows which emblems in the 11A3 dataset contain the twenty notations. Four emblems – identified by their unique Emblematica Online URLs and coded as 41, 45, 48 and 84 in my data – contain the full set with nine further emblems each containing one of the twenty notations. Since, the nine other emblems shared no other notation similarities – I confirmed this by checking using the other six notations shared by some of the four main emblems – I decided to look just at those four.
Using the URLs from the underlying data. I discovered they are essentially the same emblem, or adaptions of the same emblem. As their picturae show, in order of publication:
All four emblem picturae show a young woman (31D13), kneeling (31AA233) and protecting her head with her arm (31A253(+9161)) while hiding in a cave (25H119) from the lightning (26E2) thrown (31A2741) by a flying (31A2762) angelic figure – as evidenced by the halo (22C311) and his bird-like wings (25F3(+342)) – whom looks like a young man (31D13). The other notations indicate that she appears to have stumbled (31AA27321) and that the cave may become her grave (42E321), among other more abstract notions such as “secrecy” (52DD1(+4)) and “safety” (54D5(+4)). So, it seems that the more co-occurring notations shared by emblems, the more likely it is that they are the same emblem, or adaptations from each other or an unknown source. Though it appears they were directly copied; the bibliographic data reveals no direct connection.
These four emblems were published in four different cities – Amsterdam, Antwerp, Paris and Regensburg – by different publishers; one in the 1640s, and the other three between 1720-1745. Language-wise there is some overlap; the earliest was in French, the second latest was in Dutch and Italian, while the last was in French and German. The language of the second earliest one was not noted, but appears to be Dutch with Latin mottos, if one looks at some on the Emblematica Online site.
Step #5: Look at the other nine emblems that share only one co-occurring notation with these four. For the first few this resulted in what I expected, emblems that shared only one element with the above four, ultimately depicted differently and in different contexts. Then I looked at the emblem that shared “11A3 God’s wrath” and “25H119 cave, grotto” with the other four. Meet Emblem 69:
Instantly, it is clear this emblem belongs with the other four; however, the Iconclass notations do not reflect that fact. Why the disparity? It is because of the choices made when tagging. For example, where the four emblems were tagged with 31D13 to show there was a young woman, and with 31AA233 to show that someone was kneeling; Emblem 69 was tagged with “31D13(+54) adolescent, young woman, maiden (+ kneeling)” combining the two notations through the use of a key. Perhaps I could have addressed that by separating keys from the rest of the notations.[7] Doing so would also have fixed the difference between “25F42 snakes” tagged in Emblem 69 and the “25F42(+521) snakes (+ forward movement of animal(s))” tagged in three of the other four, thereby linking them more frequently. Though even that would not have shown how close this emblem is to the others. At this point, it seemed right to stop and re-examine my method.
Conclusion – Good Enough, but Could Be Better
It could be very easy to become frustrated by such an artifact in the data; even to the point of questioning the usefulness of Iconclass data for such a project. However, that would forget that the method did deliver what I was looking for, a set of signifiers that appeared together in the depiction of God’s wrath, even though it missed one instance. And, it works for other notations too! For example, the data for the notation “11C23 God the Father as bearded old man, usually with crown or tiara or sceptre and/or globe” includes a total of 192 emblems, shown by the “Null” column in the below graph.
There is a high rate of co-occurrence between 11C23 and “11Q71444 ostensorium, monstrance”. In Catholic and Anglican churches, ‘ostensorium’ or ‘monstance’ is the name for “vessels intended for the exposition of the Blessed Sacrament” or Eucharist (Thurston n. pag.). That an Ostensorum appears with that specific depiction of God the Father 152 times, or about 79% of the time, is quite remarkable, suggesting a incredibly strong association between the religious vessel and that depiction of God the Father. The next most common co-occurring Iconclass notation – 11M31 – only appears thirty times out of 192 (about 16% of the time). By shear comparison of frequency then, the link between the monstrance and the royal depiction of God the Father seems much more significant, and therfore worth more investigation. Thus, the method clearly helps establish questions and frame further research.
Additionally, the test on God’s wrath revealed, completely unintentionally, a bit more about Iconclass’s flexibility. The different sets of Iconclass notations are describing what is essentially the same image, but they do so in different ways. However, both are accurate descriptions of that image despite the difference. That is an intriguing observation, pointing to Iconclass’s descriptive strength, even if it causes difficulties for projects such as mine.
So, the method and the results seem to be good enough. But, of course, there are things that could be done to improve it.[8] Perhaps there could be ways to combine similar Iconclass notations in the data or in the visualisations, maybe by ignoring the keys, as I mentioned above. Alternatively, maybe the keywords Iconclass associates with each notation in the hierarchy could be used as a way to cross-check the notations with other similar ones.[9] For example, “11C23 God the Father as bearded old man, usually with crown or tiara or sceptre and/or globe” has the following keywords: Christian religion, God, beard, crown, human figure, old man, orb, religion, sceptre, supernatural, and tiara. Perhaps, in tandem with the notation co-occurrences test, a program could group all the code words for an emblem, compare them to the codes of other emblems and render a “similarity score” to researchers. Willard McCarty wrote that humanities computing, like the rest of computer science, “is principally about acting on and within the world, … studying the consequences, learning from them and devising ever better means” (189). Essentially, that is the outlook of my investigation of Iconclass notation co-occurrences. In addition to provoking questions for future research about the symbols for and surrounding God’s wrath, the experiment outlined in this post, taught me a little about Iconclass, and a lot about the method itself, prompting me to ask what worked, what delivered usable results, and what could be improved.
Endnotes
[1] Timothy Cole and Janina Sarol were kind enough to provide me with a number of co-occurrence pairs, where the first notation was the one I requested, and the second was one it appears with, along with a number saying how many times that pair occurred. I used this data as a back-up and a guide to where to start to look.
[2] Essentially, we used the process described here.
[3] Naturally, this list removes 11A3 and its subcategories, because I am interested in the other notations that appear with them.
[4] If one of 11A3’s subcategories had co-occurred with 25I1, then that subcategory’s column would appear somewhere in the set, unless it fell under the minimum co-occurrence threshold I set.
[5] I double checked that there were no instances of 25G3 that had the species of tree identified, done in Iconclass notation using parenthesis as part of the notation. There were none.
[6] “4th Co-occurrence” shows that while 11A3 and 23C311 co-occur five times, the other nineteen only co-occur four times
[7] John Brosz actually made it possible for me to do so, but I decided against it, believing it better to keep each notation distinct. I am now reconsidering.
[8] My switch to looking at the “Emblem DNA” was exactly such an attempt at improvement, specifically looking for a better visualisation, that could reveal aspects of the data in a clearer way.
[9] Iconclass lets users search for notations using the notation code or keywords. See http://www.iconclass.nl/about-iconclass/what-is-iconclass and http://www.iconclass.nl/iconclass-2100-browser for more info.
Works Cited
“Contents of Iconclass.” Iconclass. RKD, 2012. Web. 13 Jan. 2016.
“Create a Co-Occurrence Visualization Using Parameters and Sets.” Tableau Knowledge Base. Tableau Software, 24 Mar. 2016. Web. 21 Apr. 2016.
Emblematica Online. University of Illinois Board of Trustees, 2015. Web. 24 Apr. 2016.
Emblemes d’amour diuin et humain ensembleexpliquez par des vers franðcois. A Paris [Paris, Fr.]: Chez Pierre Mariette ruèe S Iacques a l Esperance, ca. 164-. Emblematica Online. Web. 24 Apr. 2016.
Guyon, Jeanne Marie Bouvier de La Motte, Herman Hugo, and Otto van Veen. Ihren Gott liebende Seele, vorgestellt in den Sinnbildern des Hermanni Hugonis èuber seine Pia desideria, und des Ottonis Vaenii, èuber die Liebe Gottes, mit neuen Kupffern und Versen, welche zielen auf das innere Christenthum, aus dem Frantzèosischen ins Teutsche èubersetzt. Regensburg, Ger.: Verlegt von Emerich Felix Bader, 1743. Emblematica Online. Web. 24 Apr. 2016.
“Iconclass 2100 Browser.” Iconclass. RKD, 2012. Web. 13 Jan. 2016.
Iconclass. RKD, 2012. Web. 13 Jan. 2016.
McCarty, Willard. Humanities Computing. Basingstoke, Hampshire: Palgrave Macmillan, 2014. Print.
Pinamonti, Giovanni Pietro. Spiegel sonder bedrogh ofte kenisse syns selfsvor oogen gestelt aen alle deught-minnende persoonen. t’Antwerpen [Antwerp, Belg.]: Jacobus Bernardus Jouret, 1733. Emblematica Online. Web. 24 Apr. 2016.
Suderman, Jan. De godlievende ziel. Amsterdam, Neth.: Henricus Wetstein, 1724. Emblematica Online. Web. 24 Apr. 2016.
Thurston, Herbert. “Ostensorium (Monstrance).” The Catholic Encyclopedia. Vol. 11. New York: Robert Appleton Company, 1911. N. pag. New Advent. Web. 24 Apr. 2016.
Finding God in the Metadata: Distant Reading and Emblem Studies by Aaron Ellsworth is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.