Stylometry and 'wordprints'

What four decades of computational authorship studies of the Book of Mormon can and cannot show.

What the Church claims

Statistical wordprint analysis shows that the Book of Mormon was written by multiple distinct ancient authors, none of whom was Joseph Smith, Cowdery, Rigdon or Spalding.

Once the evidence is in, this claim: Weak. The claim is that the analysis shows something. It does not. The apologetic studies use closed author sets and so find the voices they were built to find, and the rival Spalding-Rigdon studies fail in the same way. Weak rather than untenable, because the method itself is legitimate; it is the conclusion drawn from it that is unsupported.

The record

  • Larsen, Rencher and Layton (BYU Studies 20:3, 1980) compared function-word usage among 24 purported Book of Mormon authors with 1,000+ words each and concluded they were distinct from one another and from Joseph Smith, Oliver Cowdery and Solomon Spalding; the study was criticized (Ernest Taves; D. James Croft) and defended.
  • John Hilton's Berkeley group (BYU Studies 30:3, 1990) used 65 non-contextual word-pattern ratios with Mann-Whitney tests on 5,000-word samples, ran 325 tests on nine control authors including Twain, Cowdery and Joseph Smith, and found that eight of nine Nephi-versus-Alma comparisons gave five or more 'rejections', i.e. over 99.5% confidence of different authors; Hilton notes the method cannot handle heavily quoted text or deliberate manipulation.
  • Jockers, Witten and Criddle (LLC 23:4, 2008) used delta and nearest-shrunken-centroid methods with seven candidates and attributed most chapters to Rigdon, Spalding and Cowdery; Joseph Smith was not a candidate.
  • Schaalje, Fields, Roper and Snow (LLC 26:1, 2011) showed that naive closed-set NSC 'can produce misleading results' when texts vary in size and the true author may be absent, and that an open-set reanalysis of the Book of Mormon gave 'dramatically different results'.
  • FAIR itself cautions that 'the jury is still out' on wordprints and that results depend on the statistical assumptions.
  • The 2013 Johnson n-gram study of The Late War, though not an authorship attribution, is the other computational claim in circulation; McGuire's critique documents its base-rate and boilerplate problems.

The moving target

The Church has never endorsed or disputed a wordprint study. Stylometric arguments have run in both directions since the 1980s — some claiming to find multiple ancient authors, others naming 19th-century candidates — and no Church publication has adopted either result. The Church's position on authorship rests on revelation, not on statistics, and so is untouched by whichever way the studies fall.

The exchange

FAIR, summarizing Larsen 1980 and Hilton 1990

Wordprints show many distinct authors within the book and exclude the 19th-century candidates; Hilton's controls make the result robust.

The answer: Stylometry can measure whether stretches of text differ in style; it cannot say why. A single author writing in voices, mixing genres (sermon, war narrative, epistle) and quoting the KJV at length will produce internal differences, and the Book of Mormon's blocks differ in exactly those ways. The 'authors' tested are defined by the book's own attributions, so the design assumes what it tests. Comparing against Joseph's later edited prose, not his dictated speech, is a further mismatch. No study has tested the obvious naturalistic model: one speaker, several registers.

What they get right: Hilton's group did run controls on known authors, and the internal stylistic differentiation of the book is a real feature that any theory of composition must account for.

Schaalje et al. (2011); FAIR

The 2008 Stanford study was methodologically incapable of finding anyone but its listed candidates, and its authors did not include Joseph Smith.

The answer: Correct. The same objection, that a closed candidate list forces a winner, applies with less force but still applies to the apologetic studies, whose 'candidates' are the book's named narrators.

What they get right: The open-set method is a genuine methodological contribution and was published in the same journal.

Both cases at full strength

The skeptics' case

Every stylometric study of the Book of Mormon, believing or critical, has the same weakness: it must decide in advance who the possible authors are, and it cannot distinguish 'different authors' from 'one author in different modes'. The critical study that tried to name a 19th-century author (Jockers 2008) failed because it left Joseph out; the apologetic studies succeed only in showing that Nephi does not sound like Alma, which a novelist could have managed. FAIR's own 'jury is still out' is the right verdict. Stylometry therefore does not help the 'no one could have written it' argument.

The apologetic case

Hilton's controls and effect sizes are hard to dismiss, and no critic has published a study showing that a single 1829 dictator produces the observed internal separation. Until someone does, the claim that the book reads like several authors stands as a fact in search of an explanation.

What would change this

A study using Joseph's dictated revelations and sermons (not his edited prose) and modelling register as a variable; or a demonstration on modern single-author texts written in voices that the Hilton measures do not separate them.

Closely related

Sources

An independent project. Not affiliated with The Church of Jesus Christ of Latter-day Saints or with any organization cited here. Summaries are written here; short quotations are attributed and linked. Every page names its sources so that it can be corrected, and every page can be challenged.