Due Tuesday, October 20: Stories, Data, and Surveillance
Of all the kinds of data we examine, unstructured text is among the most interesting and difficult. Websites, social media, research, journalism, and literature contain language whose meaning depends on grammar, context, culture, irony, and knowledge of the world. Natural language processing asks how computational methods can bring structure to that language and what becomes visible—or disappears—when words become data.
Some of the earliest computational approaches were surprisingly simple. Concordances used word frequencies to study an author or text, while stylometry analyzed patterns of style and helped address disputed authorship, including the Federalist Papers. Markov used patterns in Pushkin’s Eugene Onegin to show how local probabilities could produce larger regularities. More recent methods use machine learning, sentiment analysis, and large language models, but they inherit the same basic problem: a measurable pattern is not automatically the same thing as meaning.
Simple word frequencies can still surface patterns in large texts, even though they ignore syntax and context. Sentiment analysis makes a stronger claim by assigning emotional values to language and tracing how those values change across a text. This can reveal a large-scale narrative shape, but it can also misread negation, irony, ambiguity, unusual diction, or culturally specific expression.
Your own knowledge of a novel therefore matters. In class, we will use Voyant Tools to explore what computational analysis makes visible in a familiar text—and what it misses.
This week we connect that work on language and sentiment analysis to the systems that collect and organize data. The same critical attention matters when databases organize information about people: what do the records represent, what can be inferred from them, and who can use them?
1. Read The Shapes of Stories, Section 1
Read The Shapes of Stories, Katherine Elkins (2022), Section 1 only: “The Birth of a Field.” Consider how computational approaches to narrative relate to your own experience of reading. What might an emotional arc reveal about a story, and what might it leave out?
Have your familiar novel and its clean .txt file available for class. We will draw our predictions of the novels’ emotional arcs in class, before viewing computational results.
2. Understanding Data Engineering, Chapters 1–2
Complete the first two chapters of DataCamp’s Understanding Data Engineering:
- Chapter 1: What is data engineering?
- Chapter 2: Storing data.
Focus on how data moves through a pipeline, the difference between structured and unstructured data, and the roles of SQL databases, data warehouses, and data lakes. These concepts help explain how information collected in many settings can be stored and made available for analysis.
3. Watch: How much privacy is left in an age of surveillance?
Watch PBS Horizons, “How much privacy is left in an age of surveillance?” (September 24, 2026; 26 minutes 46 seconds). Captions and a transcript are available on the page. William Brangham speaks with Byron Tau and Calli Schroeder about data brokers, vehicle tracking, facial recognition, and the connections between commercial data collection and government surveillance.
Come prepared to discuss: How can information collected for convenience or advertising become a tool of surveillance? Identify one example from the discussion and consider how collecting, storing, or combining data makes it possible.
Tuesday in class
We will draw the novels’ emotional arcs, explore TensorFlow’s word-vector projector, and play Wordle.
Due Thursday, October 22: Privacy and the Surveillance State
Two developments make large-scale surveillance possible: the mass collection of data and its storage, retrieval, and analysis through increasingly powerful databases. Information that once required human informants or targeted observation can now come from phones, browsers, cameras, and other everyday technologies.
We will consider what it means that private companies collect so much of this information and how commercial and government surveillance connect. What limits should a democratic society place on surveillance? How should we weigh claims about safety and security against harms to privacy and freedom? Even if someone feels they have “nothing to hide,” how might surveillance change what people feel free to read, say, or do?
1. Surveillance Valley
Read Yasha Levine, Surveillance Valley: The Secret Military History of the Internet (2018), the introduction and pages 151–201. If needed, download and rotate the scan for easier reading. Use the printed page numbers.
Read this as a historical account of the relationship between internet infrastructure, commercial data collection, and government power. Compare Levine’s argument with Tuesday’s PBS discussion: where do the accounts reinforce or complicate one another? Why does Levine challenge a sharp division between private and public surveillance, and how does he assess Snowden’s proposed response?
2. Privacy harms and the “nothing to hide” argument
- Neil M. Richards, “The Dangers of Surveillance”, Harvard Law Review (2013): first page only.
- Danielle Keats Citron and David Gray, “Addressing the Harm of Total Surveillance: A Reply to Professor Neil Richards”, Harvard Law Review Forum (2013).
- Robin Doherty, “Why Privacy Is Important, and Having ‘Nothing to Hide’ Is Irrelevant” (2016).
The 2013 legal readings offer arguments about the harms of surveillance; read them as contributions to that debate rather than as a summary of current law. What does Richards mean by intellectual privacy? Why do Citron and Gray argue that his account does not go far enough? How does Doherty connect privacy with democratic life?
Optional example: what metadata reveals
“What Reporter Will Ockenden’s Metadata Reveals About His Life”, ABC News (2015). This example shows how records of activity can reveal patterns about a person even without the contents of their communications.
Thursday in class: SQLZoo
We will practice querying databases with SQLZoo. What changes when individually ordinary records can be searched and combined—and why might that matter even when someone has “nothing to hide”?
Previous homework: Python and Natural Language Processing, October 13.