Deriving Insights from Historical Texts: An Exploration of Close and Distant Reading Approaches

Learning Objectives

1. Derive insights from historical speeches through close reading.
2. Use AI-assisted coding to derive insights from a historical corpus through distant reading.
3. Examine the advantages and disadvantages of close and distant reading.

Overview (see attached for complete instructions)

Close Reading portion:

Before class, students completed a writing response assignment, where engaged in a close reading of (3-5 paragraphs of their choice from John F. Kennedy’s commencement speech at American University in June 1963. Before reading this speech, students read:
(1) The assigned pages from George Herring’s The American Century and beyond; this provided essential background information for JFK’s foreign policy.

(2) A brief discussion of close reading for historians (though there was a very limited range to choose from; next time, I’ll draft my own.

Students produced a short close reading, no more than one page single-spaced. 

Distant Reading portion:

Before class, students downloaded a zip file from Canvas and added it to their Google Drive. This zip file contained the XML files and Google Colab notebook template required to complete this activity. Students worked in pairs and were guided through the Google Colab notebook, which was divided into several portions:

(1) Set-Up & Introduction to Google Colab: Students learned the basics of printing and storing text in Python, how file paths work, and linked the Colab notebook to their Google Drive.

(2) Getting Started with our Corpus: Students learned about XML files and extracted distinct elements from XML files in the FRUS Corpus.

(3) Cleaning the Files: For all files in the FRUS Corpus, students ran code to isolate the relevant XML tags and added the cleaned output to a .txt file.

(4) Tokenization, Lemmatization, Stop Words & Topic Modeling: Students tokenized and lemmatized their corpus, and also removed stop words. Students reflected on the implications of removing certain words and added custom stop words. Then, they used Gensim’s LDA model to perform topic modeling and derive insights. Students compared topics derived from one text file (JFK’s American University speech) and the larger FRUS corpus. Students also generated visualizations of topics (bar charts and word clouds) through the Python Matplotlib library.

(5) Vector Embeddings: Students used the word2vec model to generate word vectors to find words that are used in similar contexts in the FRUS corpus. Then, students compared words by calculating the ‘similarity’ between pairs of words. 

Students then practiced querying Google Gemini, which is integrated into Google Colab, to explore queries of their choice and derive more insights from the FRUS corpus.

Finally, students reflected on the activity and the insights they were able to derive from close and distant reading approaches.

Reflections

Professor Engerman: While I am wary of AI as an analytical instrument, this course offered an opportunity to “test drive” AI as a tool of historical analysis. It introduced students to a variety of historical tools and methods, so a week on AI fit right in. A number of students had trouble conducting the data-cleaning… but this served to illustrate the kinds of assumptions, generalizations, and leaps of faith that work with text as data entails. I would gladly repeat this exercise in this particular class, though would probably tinker with the close reading text; the discrepancy between Kennedy’s idealistic oratory and the messy making of foreign policy was large, making it harder to conduct an apples-to-apples comparison.

Sophia (SAIL): Before completing this activity, many students had no prior experience with coding. Therefore, guiding students through the setup and technical aspects was important, and slowly introducing Python before delving into the distant reading portion of the activity was essential. Future iterations of this activity could consider various approaches to ensuring students are set up before the start of class, as some students had not uploaded their files to their Google Drives prior to the activity, leading to a late start.

Having Gemini “vibe coding” integrated into the assignment was essential for students to test out coding on their own, even if they had no prior experience. Additionally, having Gemini as a tool to help students when they got stuck was useful, but still required additional help for some. Suggesting key phrases or terms to improve their prompts was helpful and having students share successful prompts helped the class be more engaging. Conducting this activity in pairs was useful, and students were able to collaborate and debug together. 

Students were able to make connections to examine the topics generated and the vector embeddings. Students were also able to apply their knowledge of the Kennedy administration to ask more questions through distant reading. I was pleasantly surprised by students’ discoveries when they got the chance to test out distant reading on their own. While some still struggled to debug and run their code, some especially took advantage of the lesson to investigate questions they had formulated during their close reading exercise. Seeing students get excited about future uses for distant reading was rewarding, and I hope they continue to explore this approach to historical corpora.

Readings and Resources

Brigham Young University. “Quick Guide to Close Reading,” BYU Department of History, 2024, https://history.byu.edu/close-reading#:~:text=Close%20reading%20is%20th…

Drucker, Johanna. “Why Distant Reading Isn’t,” PMLA 132:3 (March 2017): 628–35. https://www.jstor.org/stable/27037376.

Herring, George. The American Century and Beyond (2011), chap. 9 (part), p. 403-430.

Kennedy, John F. “Commencement Address at American University,” JFK Library, 1963, https://www.jfklibrary.org/archives/other-resources/john-f-kennedy-spee…

Office of the Historian, U.S. Department of State. “Foreign Relations of the United States,”
Github, accessed April 2026. https://github.com/HistoryAtState/frus

Varnum, Michael E. W., Nicolas Baumard, Mohammad Atari, and Kurt Gray. “Large Language Models Based on Historical Text Could Offer Informative Tools for Behavioral Science.” Proceedings of the National Academy of Sciences 121, no. 42 (2024): e2407639121. https://doi.org/10.1073/pnas.2407639121.