~/projects/general-conference/questions
Questions and answers
A question from the pulpit is rarely a request for information. This study takes every sentence that ends in a question mark, asks who asks and how often, whether the question is put to the audience or to the speaker, what it is about, where in the talk it comes, and whether the talk goes on to answer it.
What stands out
- 2.24 questions per thousand words, 44% of them answered in the next sentences. The rate went from 4.93 in the 1850s to 1.73 in the 2020s.
- 20% of questions are put to the audience, 26% to “us”, 6% to the speaker, 7% reported from someone else.
- The largest kind of question is about god, lord, spirit (15.87%), then lord, said, jesus (12.61%). One telling: “Is that coupled with an absolute trust that He allows some earthly suffering because He knows it will bless us, like a refiner’s fire, to become like Him and to gain our eternal inheritance?” (Donald L. Hallstrom, 2016).
- 15% of questions come in the first fifth of a talk and 17% in the last; 27% are followed at once by another question.
- Primary ask the most (2.69 per thousand words), Patriarch the fewest (1.25).
- Mark E. Petersen asks more than anyone with eight or more talks since 1950, 8.84 per thousand words; Francis M. Gibbons the fewest, 0. Orson F. Whitney's “null” (1930) is built of questions: 16 of them.
Loading the study…
Model card
- Data
- Sentences ending in a question mark, three to sixty words, in 10,629 talks of 300 words or more: 53,098 questions.
- To whom, and answered
- A question is quoted if it opens with a quotation mark or follows a verb of saying; otherwise it is put to you, to us, to myself or open, by its pronouns, in that order. It counts as answered if one of the next two sentences is close to it in embedding space (cosine 0.5 or more, all-MiniLM-L6-v2) or the next sentence opens with an answering cue (yes, no, because, the answer, it is…).
- Kinds
- K-means with ten clusters on the sentence embeddings; each kind is named by the words at least twice as common in it as outside, and illustrated by the questions nearest its centre from different speakers.
- Known failure modes
- Transcribed questions depend on the editor's punctuation, and nineteenth-century transcripts punctuate differently. The answered test is a proxy: a question answered three paragraphs later counts as unanswered, and a question echoed rather than answered counts as answered. Cluster names are word lists, not categories a reader would choose.
Methods and data for the corpus as a whole.