~/projects/general-conference/compared-with-byu/distinctive

What only Latter-day Saints say

General Conference and BYU Speeches, 30 million words, set against 7.6 million words of other preaching and oratory: every papal encyclical in English from Leo XIII to Leo XIV, 780 of Spurgeon's sermons, Wesley's Sermons on Several Occasions, and every presidential inaugural address. The question is which words belong to this pulpit and to no other, and which it shares with the rest of Christianity.

What stands out

Sin and grace are the words that mark the other pulpits. Against everything else, the two strongest marks of the outside corpora are “sin” (1448.9 per million outside, 250.8 here) and “grace” (1095.2 outside, 97.6 here); Spurgeon and Wesley say “sin” seven times as often as conference does.
The Latter-day Saint pulpit talks about a church and a people. Its strongest marks against all four outside corpora are church, people, young, saints, day, prophet, gospel, priesthood, years, president: an institution, a lineage of prophets, families, and testimony, where the others speak of the soul, mercy, blood and the sinner.
Against the encyclicals the gap is social thought. The papal documents are marked by human, christian, catholic, social, apostolic, dignity, divine, mystery; conference and BYU against them by lord, me, said, know, day, come, am, the first person and the spoken story.
Held to the same decades, Brigham Young's Utah and Spurgeon's London were preaching different sermons. The Journal of Discourses is marked by people, earth, saints, kingdom, nations, joseph, brethren; Spurgeon by christ, sin, love, soul, grace, must, even.
92 common words appear almost nowhere outside. 65 are names and places (byu, brigham, nephi, utah, kimball, moroni); 27 are vocabulary: tithing, campus, dollars, quorums, resurrected, seventies, dad, tobacco, graduate. That vocabulary made up 2.38 words per thousand of conference in the 1850s, peaked at 4.86 in the 2000s, and stands at 3.85 in the 2020s.
What is shared is the New Testament. The words common to all five religious corpora and absent from the inaugurals begin christ, jesus, gospel, sin, paul, sent, kingdom, heavenly, sins, scripture.

Loading the study…

Model card

Data: Conference (10,803 talks, 1839–2026) and BYU (2,546 speeches); 216 encyclicals from vatican.va (1878–2026); every fourth volume of Spurgeon's Sermons from the Christian Classics Ethereal Library (780 sermons, 1855–1915); Wesley's Sermons on Several Occasions from the same library (140 sermons, undated in the edition); and 55 inaugural addresses from Project Gutenberg (1789–2005). Only aggregates are published.
Measure: Weighted log-odds ratio with an informative Dirichlet prior (Monroe, Colaresi and Quinn, 2008): for each word, the log-odds of its use in one corpus against the other, shrunk toward the pooled rate by a prior of 2,000 pseudo-words, divided by its standard error. Lowercased words of two letters or more, contractions and a short stop list removed, words with fewer than ten occurrences in the pair ignored. The z-score favours frequent words with large differences, which is why common words lead the lists.
Only-here and common-core lists: A word is “only here” if it reaches fifteen per million in conference and BYU together and occurs at most twice in the outside corpora; a name if it is capitalised in four fifths of its uses. The common core is every word at forty per million or more in each of the five religious corpora and at eight or fewer in the inaugurals. The decade line counts tokens from the only-here list per thousand words of each conference decade.
Known failure modes: The outside corpora differ in century, genre and medium as well as in church: encyclicals are written documents, Spurgeon and Wesley preached in the eighteenth and nineteenth centuries, the inaugurals are civic oratory. The matched pairs control for some of that and the rest should be read with it in mind. Spoken transcripts carry words of the room (“today”, “said”, “me”) that written documents do not. Optical character recognition in the older Conference Reports and the CCEL texts adds a little noise at the tail of every list.

Methods and data for the corpus as a whole.