The research I am doing presently uses visualizations to show latent patterns that may be detected in a set of poems using computational tools, such as topic modeling. In particular, I\u2019m looking at poetry that takes visual art as its subject, a genre called ekphrasis, in an attempt to distinguish the types of language poets tend to invoke when creating a verbal art that responds to a visual one. Studying words\u2019 relationships to images and then creating more images to represent those patterns calls to mind a longstanding contest between modes of representation\u2014which one represents information \u201cbetter\u201d? Since my research is dedicated to revealing the potential for collaborative and kindred relationships between modes of representation historically seen in competition with one another, using images to further demonstrate patterns of language might be seen as counter-productive. Why use images to make literary arguments? Do images tell us something \u201cnew\u201d that words cannot?<\/p>\n
Without answering that question, I\u2019d like instead to present an instance of when using images (visualizations of data) to \u201csee\u201d language led to an improved understanding of the kinds of questions we might ask and the types of answers we might want to look for that wouldn\u2019t have been possible had we not seen them differently\u2014through graphical array.<\/p>\n
Currently, I\u2019m using a tool called MALLET to create a model of the possible \u201ctopics\u201d found in a set of 276 ekphrastic poems. There are already several excellent explanations of what topic modeling is and how it works (many thanks to Matt Jockers<\/a>, Ted Underwood<\/a>, and Scott Weingart<\/a> who posted these explanations with humanists in mind), so I\u2019m not going to spend time explaining what the tool does here; however, I will say that working with a set of 276 poems is atypical. Topic modeling was designed to work on millions of words, and 276 poems doesn\u2019t even come close; however, part of the project has been to determine a threshold at which we can get meaningful results from a small dataset. So, this particular experiment is playing with the lower thresholds of the tool\u2019s usefulness.<\/p>\n When you run a topic model (train-topics) in MALLET, you tell the program how many topics to create, and when the model runs, it can output a variety of results. As part of the tinkering process, I\u2019ve been working with the number of topics to have MALLET use in order to generate the model, and was just about to despair that the real tests I wanted to run wouldn\u2019t be possible at 276 poems. Perhaps it was just too few poems to find recognizable patterns. For each topic assignment, MALLET assigns an ID number to the topic and \u201ctopic keys\u201d as keywords for that topic. Usually, when the topic model is working, the results are \u201creadable\u201d because they represent similar language. MALLET would not call a topic \u201cSea,\u201d for example, but might instead provide the following keywords:<\/p>\n blue, water, waves, sea, surface, turn, green, ship, sail, sailor, drown<\/p><\/blockquote>\n The researcher would look at those terms and think, \u201cOh, clearly that\u2019s a nautical\/sea\/sailing\u201d topic, and dub it as such. My results, however, on 15 topics over 276 poems were not readable in the same way. For example, topic 3 included the following topic keys:<\/p>\n 3 0.04026 with self portrait him god how made shape give thing centuries image more world dread he lands down back protest shaped dream upon will rulers lords slave gazes hoe future<\/p><\/blockquote>\n I don\u2019t blame you if you don\u2019t see the pattern there. I didn\u2019t. Except, well, knowing some of the poems in the set pretty well, I know that it put together \u201cLandscape with the Fall of Icarus\u201d by W.C. Williams with \u201cThe Poem of Jacobus Sadoletus on the Statue of Laocoon\u201d with \u201cThe New Colossus\u201d with \u201cThe Man with the Hoe Written after Seeing the Painting by Millet.\u201d I could see that we had lots of kinds of gods represented, farming, and statues, but that\u2019s only because I knew the poems. Without topic modeling, I might put this category together as a \u201cmasters\u201d grouping, but it\u2019s not likely. Rather than look for connections, I was focused on the fact that the topic keys didn\u2019t make a strong case for their being placed together, and other categories seemed similarly opaque. However, just to be sure that I could, in fact, visualize results of future tests, I went ahead and imported the topic associations by file. In other words, MALLET can also produce a file that lists each topic (0-14 in this case) with each file name in the dataset and a percentage. The percentage represents the degree to which the topic is represented inside each file. I imported the MALLET output of topics and files associated with them into Google Fusion Tables and created a dynamic bar graph that collects file-ids along the vertical axis and along the horizontal axis can be found the degree that the given topic (in this case topic 3) is present in the file. As I clicked through each topic\u2019s graph, I figured I was seeing results that demonstrated MALLET\u2019s confusion, since the dataset was so small. But then I saw this:<\/p>\n
![here<\/a>.]\n](\"http:\/\/www.lisarhody.com\/wp-content\/uploads\/2012\/04\/gazer-spirit_visualization.jpg\"){kind=link}