For this assignment, I chose the book The Adventures of Sherlock Holmes by Arthur Conan Doyle. While I have never read the collection of stories, I did watch the movie starring Robert Downey Jr., so I know the gist of the plot and characters. After downloading the raw txt file (.txt) of the book, I experimented with text analysis using Voyant Tools and Google’s Gemini. On Voyant Tools, I examined the word frequency and trends of words in different segments of the book after removing common stop words, such as said. An interactive visualization of the word cloud, line graph, and word count summary that were generated is shown below.
I was not surprised at the results when I first saw this data. It lined up with what I expected to see from the book. I appreciate how easy it is to use Voyant Tools. It’s very transparent in showing how often words appear and where they appear in the text; however, something that knocks Voyant Tools a little bit is that it doesn’t interpret these patterns. It doesn’t explain the meaning behind these words, leaving out important context to explain the value of the word.
I then started my experiment with Gemini using an AI-based distant reading approach. I asked the tool if it could “visualize the top 30 words as a word cloud.” It was unable to generate one and instead generated Python code to use to create a word cloud yourself on an external website. It did, though, list the most likely frequently used words, and it is very vague. Unlike Voyant tools, it doesn’t explicitly show how and where these words show up in the book, and the words it produced are not all the same compared to Voyant Tools. This puts the validity of Gemini in question.
In the age of AI, it is important to remain skeptical of how computational tools interpret data. People have been trained to eat up answers and clarifications AI presents, disregarding where the information came from and its accuracy. In conclusion, it is important to fact-check if the information you are giving is factual, especially when it comes to AI.
Super interesting idea to prompt AI to try and create its own word cloud Jackson. The direct comparison would’ve been interesting. I’d like to see what that Python code would produce. I imagine it wouldn’t be nearly as interactive as the Voyant tool but still would be cool to compare.