Category Archives: Sherlock Holmes Text Analysis

When I first used Voyant I felt very much like I do when I walk into a record shop; excited, completely overwhelmed by the amount of content, and uncertain of where to begin. To make things a little easier on myself I limited my exploration to “A Study in Scarlet” rather than the entirety of the Sherlocks Holmes corpus. I began by editing the word cloud and common word tools so that they would no show common words such as “the, a, and, etc”, then I looked at the results provided by the tools and realized that they did little to increase my understanding of the text beyond point out the obvious, such as the frequent presence of the word “said” in a story that involves dialog and narration. Next I proceeded to attempt to create a custom collection of tools to analyze the text. The keyword here is attempt, because my effort proved nigh-fruitless. I decided to keep the corpus reader and the word trends tools, then I added several other information visualizers, such as the bubblelines and knots. Unfortuantely I could not discern how to work these visualizers. After failing to display anything for several minutes, attempting to access the help page and finding it empty, I decided to focus on the word trends tool. Initially I was uncertain of what to look for, then I thought “how does one go about understanding a situation/do detective work?”. The answer that I decided upon was through the use of one’s senses and thought processes. I set up a word trend graph showing results for “saw”, “heard”, “felt”, “tasted”, “smelled”, and “thought”, then proceeded to exmanine the graph in an attempt to see which senses Sherlock seemed to use most often and if there was any correlation between and subsets of the terms. Tasted and smelled both returned no results, so I ignored them. On average, saw and thought returned the most results and there seemed to be somewhat similar and overlapping changes in the frequency and locations of the appearences of saw and felt. Almost every line approached zero near the center of the graph. I proceeded to add “sherlock” to the graph, and the decrease in the frequency of the appearence of the keywords corresponded with a lenghty disappearence of the word sherlock from the text.

When playing with Sinclair and Rockwell’s Voyant, I decided to take a look at the entire Sherlock Holmes corpus in order to view trends that spanned across the collection. I tampered with the different formats of data visualization (the most interesting of which was the bubbles, which made a shrill, high pitched noise as it went through the entire text word by word) , but the most useful one in my perspective was the default Cirrus (word cloud) format.

With Cirrus, you can see the giant word cloud of the most frequently used words, with the size corresponding to the number of occurrences in the Sherlock Holmes corpus. After filtering out the common and irrelevant words (by selecting the Taporware option in the “Stop Words List”), the user gets left with a more relevant view of the text. One of the first things I noticed was the relatively large word “man.” Subconsciously going back to our class discussion regarding Arthur Conan Doyle’s perspective of women (a la “Scandal in Bohemia”), I found it interesting that the word “man” was so much more prevalent than any word pertaining to a female. The largest I could find was “lady,” which at 176 instances was dwarfed by both “sir” (at 323 instances) and “man” (weighing in at a whopping 902 instances).

The Cirrus view also maintains the Words in Documents window, which allows the user to see exactly how many times the word occurred in each document. I found it interesting that, out of 36 stories, the word “lady” did not come up in 12. I feel like this reveals a little bit more about Arthur Conan Doyle’s perspective on women—although he may respect them (by making one as a protagonist who outsmarts Holmes in “Scandal in Bohemia”), he still grew up in a patriarchal society that was mainly concerned with reading about men instead of women.

There was one thing I would have liked to see, though, which is having more than one word on the Word Trends graph. I’m sure Voyant has this capability, and I’m relatively sure I’ve gotten it to work before, but for some reason this time tampering with it did not yield the same results as the last time.

I’m having a lot of trouble getting beyond a superficial understanding of the collection of stories. If you use the summary tool, you can get an idea of what each story is about. For instance, the top words from a Scandal in Bohemia are photograph, king, majesty, Adler, and Irene. If you click on a story, you can perhaps get an even better idea by using the Words in Documents, Keywords in Context, and Word Trends tools together.

So, I’m definitely understanding these tools as ways to practice computer assisted reading—if I’m understanding it correctly as a way to more efficiently scan a document, but focusing on its keywords.

But, I have not been successful in making any headway in this “screwing around” methodology. There are perhaps several reasons. First and foremost, I don’t think I understand the tools and how to use them effectively. Mostly what keeps coming to my mind is, I don’t know what I’m looking for… which I realize is exactly the point of Ramsay’s essay, so I guess I don’t fully buy into his idea.

I suppose, to some degree I’m “looking for” an interesting image that makes me ask a significant question, or leads me to some kind of conclusion that brings some kind of significance to what I’m doing. But I haven’t seen anything yet, except the summarization quality of the tool, which allows me to get an idea of what the stories are about.

I thought that, perhaps, one of the more visual tools would be more engaging, so I tried both the knots tool and the collocate clusters tool. I found both these tools to be less than helpful in my understanding of the texts. When I tried to click on the various segments of the knots, the screen would reload, and the lines would draw themselves over and over again. With the collocate clusters, I liked that I could visually see what was apparent from the summary tool, but it didn’t deepen my understanding any.

In summary, I will be the first to admit that I have been hindered by a lack of interfacing skill in my ability to arrive at any conclusions from screwing around on Voyant; however, as I was screwing around, I couldn’t help but think that if I had something specific for which I was looking for support, I could definitely use these tools to help me find it.

It took me a while to settle in on what I wanted my searches to be focused on. For about 30 minutes I was just typing in arbitrary words like “clue”, “fight”, “gun”, etc to try and find the faster-paced parts of the story. I will make a note that the word trend function makes it easier to find a specific part of a book; like a deduction scene, chase scene, romantic scene, etc.. I finally tried trying in the names of each of the character to see how often they came up in conversation. Interestingly enough, the word ‘Holmes’ showed up more than the word “Sherlock”. Also, the marked incline of ‘Holmes’ after the 4^th section might be symbolic of Sherlock Holmes getting closer to solving the mystery. On the bottom is Watson’s name…he doesn’t come up nearly as often as the main character (naturally), and is absent during a majority of the 2^nd half of the section.

I then switch over to the whole corpus view and did the same name search but limited it to only “Sherlock”, “Watson”, and a character from Study in Scarlet named “Gregson”. This way, I would be able to see which of the stories mentioned any particular character a lot and which mentioned them only a few times. As you can see from the ‘Sherlock’ and ‘Watson’ graphs below, each character is motioned a varying number of times; there are clear upper and lower boundaries. I included the last “Gregson”  example as a means of displaying how this application can help you find even the most specific characters in a story. I ran across ‘Gregson’ in the first story I read and then wanted to see if he showed up in any other stories in the corpus. The last pane shows you that Gregson only shows up in Study in Scarlet.