I chose Robinson Crusoe as my text because I read it recently for an English class. I then decided to use Voyant and Gemini for my textual analysis. I tried several tools on Voyant like TextualArc, Bubbles, RezoViz, Knots, and Cirrus. Some of them, like RezoViz, which needs the program to identify locations in the text, didn’t work so well. While others like Knots, which draws knots on lines according to frequency, were visually very confusing and not very clear to the viewer. Even though though some of them didn’t work so well, there was a lot of variety in Voyant and all of these tools would work well for different purposes. Looking at bubbles in the illustration above, for example, I am able to understand the setting and general content of the novel. Words like shore, ship, boat, island give a clear image of where the story takes place. And the repetition of the word thought shows how self reflective the novel is.
Then I chose Gemini for my second visualization experiment, and according to Anastasia Salter’s instructions, gave it this prompt: “Can you visualize the network of character relationships in this text?”
I thought Gemini did a great job with this image. As a person who is familiar with the text, i can perfectly approve that this is a very clear and true relationship map between the characters. The visuals also are very helpful, similar roles are color coded and the background fits with the setting of the story. I honestly did not expect Gemini to do a good job at all. Most often AI tools get confused by prompts and then refuse to listen to feedback. But would I trust AI if I didn’t actually know the text? No. I mostly think this is a freak accident. But I know a lot of people who would trust AI to give them information without questioning it and I think that is problematic. I think responsible AI usage should be taught to everyone, explaining how AI works and where it gets its information, what are the environmental consequences, and how truthful its information is.
I enjoyed your analysis of these tools! I was intrigued by the problems you found with the Knots tool. I wonder if it would be visually digestible with a smaller text pool. I worked with Spyral instead of testing an AI model, so I found your experience with Gemini interesting as well. AI is getting better at creating these analyses, especially compared to the example Professor Mason showed in class, so perhaps with another prompt it could produce a different type of display that shows if it truly understands the novel’s text.