Method · stylometry
Burrows' Delta
Measures how alike two texts are in their use of the most common words, so you can tell which texts were probably written by the same hand or in the same language.
Questions it answers
- Which pages or texts were written by the same scribe?
- Do sections of a text group by subject, by hand, or by language?
- Is an unknown text closer to one candidate corpus than another?
How it works
Take the most frequent words across all the texts. For each text, turn each word's frequency into a z-score against the whole corpus. Delta between two texts is the mean absolute difference of those z-scores. Small Delta means similar style.
Δ(a, b) = (1/n) Σ | za(wᵢ) − zb(wᵢ) |
Assumes
- Texts long enough to give stable word frequencies
- The same transcription and word boundaries throughout
Watch out for
- Topic can masquerade as authorship; control for subject
- Results depend on how many words you keep
Where it comes from
Built on
Function-word frequenciesMosteller and Wallace, 1964
Standard score (z-score)Theory · statistics
Burrows' DeltaBurrows, 2002
Led to
Cosine DeltaSmith and Aldridge, 2011
ZetaRelated: distinctive words
Code Python 3.11
from mysteriesdept.methods import burrows_delta
tokens = load_dataset("voynich-token-table") # one row per word
pages = tokens.groupby("page")["word"].apply(list)
delta = burrows_delta(pages, n_words=150) # pages × pages matrix
nearest = delta.nearest(k=1) # each page's closest page
Runs and results 18 runs · 6 shown
Applied to
Voynich ManuscriptPages group by scribe and by Currier language more strongly than Latin texts group by author.
3 datasetsExploration split, Latin genre texts, ZL3b transcription
References
- Burrows, J. 2002. 'Delta': a Measure of Stylistic Difference and a Guide to Likely Authorship. Literary and Linguistic Computing.
- Smith, P., Aldridge, W. 2011. Improving Authorship Attribution: Optimizing Burrows' Delta Method. Journal of Quantitative Linguistics.
- Evert, S., Jannidis, F., Proisl, T. et al. 2017. Understanding and explaining Delta measures for authorship attribution. Digital Scholarship in the Humanities.