This project tracks associations between the languages of consumer culture and ethical practice in Early Modern English texts.
A photo of Josiah Child’s 1693 A New Discourse of Trade from the David M. Rubenstein Rare Book & Manuscript Library, Duke University.
Is there a right type and amount of consumption? Building on the recent developments in the historiography of the Early Modern development of the concept of “consumer” and “consumerism,” this project tracks associations between the languages of consumer culture and ethical practice. Working with undergraduate and graduate student teams from Duke, BSC, and UC Irvine, Astrid Giugni (English, Duke) and Jessica Hines (English, Birmingham Southern College) analyzed the approximately 70,000 Medieval and Renaissance digitized texts made available by EEBO-TCP through the use of Natural Language Processing methods.
Final team poster detailing their methodologies and findings. PDF available here.
This poster by our Summer 2020 team outlines their methodological approaches to analyzing the language of consumption and consumer culture. The team used word embeddings to track the relationships between consumer-related words, and they began analyzing the gendered and antisemitic valences of such terms. For more technical details see the below video:
t-SNE plot word cluster representing the relationship between 30 consumption terms.
This is a t-SNE plot the team used to track the relationship between a cluster of consumption terms. Words clusters that are similar to each other will appear closer together. The large scatter plot visualization is of 30 consumer-related words.
t-SNE plot of four words: consume, possess, desire, luxury.
In this t-SNE plot, the team chose four words. “Consume” and “luxury” are close together, meaning that they are more similar and appear more often together.
Final team poster detailing their methodologies and findings. PDF available here.
Our Summer 2021 team’s poster outlines how they used sentiment analysis to track cultural response to key trade goods. They also used topic modeling and word embeddings to understand religious and philosophical attitudes toward trade. For more technical details see the video below:
Image grab from project.
Early Modern text analysis presents a unique challenge: spelling is not yet standardized nor is typeface. To analyze texts with greater accuracy, the team used the program VARD to normalize spelling.
A text with a list of possible options for normalizing the spelling of the word “corporal.”
This process wasn’t always straightforward. Some Early Modern word forms are so different from our current spelling that it can be hard to tell what words are without context. The team had to identify problem terms to further analyze.
2021 Topic Modeling Sample
Once spelling was normalized, the team used topic modeling to organize words into groups. That allowed them to see what words were closely related to each other or appeared with frequency near each other.
Heatmap 1580-1584
To understand relationships between specific consumption-related terms in the corpus, the team used a word embedding algorithm to map words into vectors. This technique allowed the team to find how closely related words are by computing their cosine similarity—a measure of how close their meaning is within the corpus. This heatmap visualizes these relationships over a predetermined set of words in EEBO-TCP texts from 1580 to 1584.
Box Plot Sentiment Analysis
The team collectively decided to try to apply sentiment analysis to test its limitations as well as its potential. Sentiment analysis is a common classification tool that assigns a “sentiment” score to a text (on a positive to negative scale). While it is a blunt tool, sentiment analysis can be helpful in early explorations of a large number of texts. This box plot visualizes the sentiments expressed in a ten-word window surrounding target “consumables” (beer, gold, silver, tobacco, and wool).
Distribution Graph of Sentiment Analysis
While the box plot above gives a clear, high-level overview of the sentiments about the given target consumables, this visualization gives a more granular view of the distribution of sentiments in the same texts. The team organized the texts into two clusters, labelled “religion” and “religion + philosophy.” The division between the clusters was one of convenience, rather than of conceptual distinction, as the team attempted to organize texts based on whether they made use of specific, explicit references to “philosophical” authors, such as Plato, or terms, such as megalopsychos.
3D visualization of the co-occurrence of the terms “corruption” and “monopoly” in EEBO from 1450-1700.
The students who are currently working with us on this project are expanding the analysis to include an exploration of texts dealing with the East India Company and the development of the use and meaning of the term “monopoly.” This visualization charts which texts mention both the terms “corruption” and “monopoly” from 1450 to 1700.
Image plotting the frequency of “corruption” in EEBO-TCP.
Image plotting the frequency of “monopoly” in EEBO-TCP.
To help us understand better the context for the co-occurrence of the terms “corruption” and “monopoly,” the team also plotted the frequency of each term individually.
Faculty: Astrid Giugni and Jessica Hines, supported by Rhodes Information Initiative at Duke, the Charles B. Vail College Fellows Program at Birmingham Southern College, and Duke Bass Connections.
Undergraduate members: Donald Pepka, Andrew Vas Scofield, Albert Sun, Daisy Zhan, Shuba Prasadh, Abhishek Devarajan, Serena Ivery, Aimi Wen, Victoria Terry, Leo Proctor, Madison Blair, Amy Weng, Heidi Smith, Danila Reznichenko, Ioana Lungescu, Charlotte Lim, and Theodora Harmsworth.
Graduate members: Meghan Woolley, Emma Davenport, and Chris Huebner.
Duke University, Birmingham-Southern College
2020 - present
Consumption, ethics, corruption, mercantilism, greed, trade, printing history