Method 1- Voyant Tools
First, I uploaded my cleaned (deleted extra licensing information) Dracula .txt file into Voyant. One thing I noticed about the Cirrus word cloud is that Dracula’s name does not appear. I did a little research, and this actually makes sense since characters often refer to Dracula indirectly (like “Count” or just “he”) instead of saying Dracula. This leads to “count” appearing in the cloud, but not Dracula.
Method 2- Gemini
Next, I followed Anastasia Salter’s AI-assisted distant reading method using Google Gemini AI. I asked it to help with several specific distant reading tasks to see if it could output a similar result to the word cloud that Voyant produced.
I began by sending it the .txt file. Then, I gave it a few requests. For example:
Even without me asking, Gemini applied a standard list of English stop words, similar to Voyant .
However, when I first asked Gemini to visualize the list of words in a word cloud, it told me it could not do so and that it could only generate code for me to then run in a local Python environment.
After a few more requests, it would even tell me that it was directly creating a visual for me, but then still just send me code.
For example:
After several messages of explicitly telling Gemini that I was not at all interested in the code, and that I wanted a visual word cloud, it finally produced this image:
Looking at this word cloud and comparing it to the one generated by Voyant, I was honestly surprised by how similar they were. There were some small discrepancies, but I believe these just come from the slight differences in stop words used. Overall, Gemini did a successful job; it just took much more specific instructions along with more trial and error.
What this says about the “Age of AI”
Using Voyant and Gemini showed that different tools can produce similar results, but the way in which they reach them is different. While Gemini is capable of doing similar analysis of word frequencies, it requires more specific prompts and experimentation to get there. So, in the “Age of AI,” this shows that AI tools can definitely be powerful for text analysis, but only if users understand how things like preprocessing and prompt wording shape the output.
I also got Python code as a response to asking Gemini to generate a word cloud. I didn’t know if I continued to ask it for one, it would eventually create one. I guess it goes to show that generative AI can only be as detailed as your prompt. Having a surface-level knowledge of the subject you are using AI for produces the best results.
It is truly interesting to consider that Gemini can yield such different results based on how instructions are phrased. I think you make a great point in saying that AI requires more specificity when trying to get the exact results that are wanted. I have had to be hyper-specific at times when playing with LLMs. Overall, I think your text analysis of Dracula was really cool, and I appreciated how you displayed your troubles with Gemini. I like how you investigated the absence of the word “Dracula” as well.