Session 9
Data analysis: text corpora
Hodel Taught by Tobias Hodel.
In this session
- Discussion of (Underwood 2014) — search as an unexamined research practice
- Building a text corpus from the articles in your Zotero library
- Hands-on: Voyant Tools
From library to corpus
Your Zotero library holds PDFs. A corpus is something else: a deliberately bounded, documented set of texts, in a form a machine can read.
- What makes a corpus a corpus — boundaries, and a statement of what is in and what is out
- Getting plain text out of PDFs, and what the extraction destroys
- Cleaning: headers, footnotes, hyphenation, running heads, OCR noise
- Documenting the corpus, so that someone else could rebuild it
Distant reading, examined
Underwood’s argument is that the practices we consider neutral — searching, sorting, filtering — are already interpretive, and that we adopted them without theorising them.
- What does a keyword search assume about the relation of words to concepts?
- What does a word-frequency list actually tell you?
- When is “distant reading” a method, and when is it a rhetoric?
Hands-on: Voyant Tools
- Upload your corpus
- Cirrus, Trends, Contexts, Correlations — what each panel actually computes
- Stopwords: build your own list, and notice how much the results move
- Compare two sub-corpora — by decade, by language, by sub-topic
- Export a visualisation and a link for your project site
Then the hard question, in plenary: did any of this change what you think about your material? Negative answers are results too.
The DH Lab moves the same kind of analysis into notebooks (Colab / Noto) with topic modelling.
Reading for Session 10
Craig, Hugh. 2025. “The geography of tragedy: mapping allusions to places in a set of Shakespearean plays.” Digital Scholarship in the Humanities 40 (3): 747–761. https://doi.org/10.1093/llc/fqaf055 (open access, CC BY 4.0)
Craig collects every allusion to a place in the dialogue of twenty-two early modern English tragedies — eleven by Shakespeare, eleven not — and maps them to look for patterns across the set, by author, and play by play.
Read it as a complete workflow, not only as an argument about Shakespeare. The whole chain is published and you can inspect every link of it:
- the TEI-encoded source texts, with a metadata file recording which early printed editions they derive from — https://doi.org/10.5281/zenodo.14504814
- the map and the data behind it, in the TLCMap workbench — https://tlcmap.org/layers/2246
Questions to bring:
- What counts as “a place” in this study, and who decided?
- What did the TEI encoding have to settle before any mapping was possible?
- The article finds the densest concentration of place names in the eastern Mediterranean. Is that a finding about the plays, or about the sample?