Skip to content
james_tavita

~/projects/byu-speeches/style

Style fingerprints

Strip away what a speech is about and what remains is how it is said: the function words, the length and variety of sentences, the pronouns, the punctuation, the habit of citing. This study gives every speaker a fingerprint from those traits alone, maps them, and asks the sociological question: when someone joins the Twelve, does their voice move toward the Twelve's?

Unsupervised70 speakers1,971 speechsStyle map ~/byu-speeches

What stands out

  • On campus, ordination shows no measurable effect. Only 9 apostles have enough BYU speeches on both sides of their ordination; the mean change is 0.341 (p = 0.5703), nothing to report.
  • The first axis of style is sentence length; the second is word length. word length, long words, he drive the horizontal axis and the, semicolons, shall the vertical; together they explain 28% of the variance between speakers.
  • Sentences got shorter. From 21.19 words in the 1970s to 19.99 in the 2020s, while second-person address went from 1.69 to 1.77 per hundred words.
  • Each calling has a style. BYU faculty: less you, less second person; BYU leadership: more also, less third person; First Presidency: more here, more of; Presiding Bishopric: less word length, more thine.
  • The auxiliaries speak in the third person feminine and the second person; the Twelve use the first person singular least of any body.
Loading the study…

Model card

Data
1,971 speechs of 300 words or more; 70 speakers with five or more.
Features
Rates of 190 function words; mean and standard deviation of sentence length; mean word length; type-token ratio on the first thousand words; share of words of nine letters or more; commas, semicolons, colons, questions, exclamations, dashes and quotation marks per hundred words; first-person singular, first-person plural, second-person and third-person pronoun rates; scripture citations per thousand words; words per paragraph. No content words.
Map
Features standardised across talks, averaged per speaker, then the first two principal components.
Convergence
For each apostle with three or more talks before and after the year of ordination to the Twelve, the Euclidean distance in standardised feature space from each talk to the mean of talks by other members of the Twelve within five years. Change is the mean after minus the mean before; a paired Wilcoxon signed-rank test across apostles.
Known failure modes
Style drifts with era, so a speaker's position partly reflects when they spoke; the convergence design compares each talk to the Twelve of its own time to control for this. Transcription conventions (the Journal of Discourses reporters, the Conference Report's OCR) shape punctuation and sentence boundaries.

Methods and data for the corpus as a whole.