A Change of Genre in the Prophetic Word

Abstract. The companion study established that of the Ellen G. White citations in a century of three official Seventh-day Adventist periodicals, 83.8–89.2% — in every channel and every period — stand near none of the doctrinal themes examined. This study examines that residuum, on the same corpus (Review and Herald / Adventist Review, Ministry, Curierul Adventist; 74,520 citations in the nine analysed cells), with the same detector and null model, in three mutually checking cross-sections. (1) Composition: the residuum is the terrain of the devotional book corpus (Steps to Christ, The Ministry of Healing, The Desire of Ages), while the doctrinal side draws on the visionary-historical set (White's own periodical articles, Early Writings, The Great Controversy). (2) Registers: the residuum was mapped by embedding-based discovery, an explicit lexicalizability filter, and split-half validation. Nearly a quarter of it belongs to five dictionary-measured registers (dietary counsel, citation apparatus, book trade, international distribution, Sabbath School lesson schedule); the theological registers (sanctification, the great controversy frame, parenting) cannot be captured by dictionaries and were therefore measured with embedding centroids. Dietary counsel is the only register that significantly attracts the White citation in every channel and every period — more reliably than any core theme of the sanctuary complex. (3) Form: the strongest movement of the century is not thematic but generic, and it occurs in two stages. First, at mid-century, within running prose, the voice of the White citation gives way to the name: the share of invocations carrying no quotation block rises from 16–25% to 51–57% in all three papers. Then, after 1990, the name moves from running prose into the note apparatus — but only in the American professional journal: in Ministry 1.9% of citations stood in a note apparatus in 1920–1945 and 55.8% after 1990; in the Romanian paper the figure is 7.1%. The prophetic word did not disappear; it changed genre — from voice to name, from name to footnote. We read the results as the formal imprint of the routinization of charismatic authority.

Keywords: citation form, routinization of authority, residuum analysis, lifestyle norms, religious press, Seventh-day Adventism, Ellen G. White, embedding-based clustering, lexicalizability

1. Introduction

1.1 The question

The companion study measured when the White citation stands beside the doctrinal themes of the sanctuary complex.¹ Its most robust compositional finding was negative: 83.8–89.2% of citation environments carry none of the eleven themes examined. This study turns that residuum into a question: if it does not stand beside sanctuary doctrine, where does the White citation stand in the official church press — beside which topics, in what kind of text, and in what form?

The question is not a matter of leftover curiosity. If the great majority of White invocations are not doctrinal proof, then the actual, everyday working of the authority is carried precisely by the residuum — by what the proof-text debate, the central object of the normative literature, has never measured. The question is more general than the first study's, and the method must be more general too. The first study searched for pre-specified themes. Here we proceed in reverse: first we let the residuum show its own clusters, and only then do we operationalize as a measurable theme whatever survives that translation. Between the two we insert an explicit filter that states what can honestly be measured with a dictionary and what cannot.

1.2 How does this relate to the first study?

The two studies use the same corpus, the same citation detector and the same null model, but ask different questions. The first tested a single, pre-formulated hypothesis: does the White citation pull toward the doctrinal load-bearing point, the sanctuary complex? This study begins where the first study's answer ended, and first discovers, then measures, what occupies the remainder.

The decisive difference, however, is not scope but angle of view. By construction, the first study could see only which theme stands near the citation. This study can also see what kind of text it stands in — argument, advertisement, curriculum or note apparatus — and in what form: does the paper reproduce White's text, or merely name her? Its main result, the change of genre in the prophetic word, is a claim about genre and form that the first study's method could not have detected. For the same reason, this study can also measure the reach of the first study's method: it shows which registers can be captured by a dictionary and which cannot.

In one respect the two studies are identical. Both measure use through placement: the first study's premise is that citation is a church practice and that the location of the citation is diagnostic. Neither measures directly the rhetorical function of the individual citation — whether it proves, illustrates, disputes or sells; that is decided by the manual function coding now under way. This is why this study speaks of registers and form: the two words name what the measurement actually sees.

1.3 The answer in four sentences

The residuum is not empty but differently occupied: in its source works it is the terrain of the devotional canon, nearly a quarter of it (23.4%) belongs to five dictionary-measured registers, and further layers of it can be identified by embedding-based measurement.

The most stable terrain of authority use is not doctrinal but concerns everyday life: dietary counsel is the only register that significantly attracts the White citation in every channel and every period (lift 1.7–3.5).

The strongest movement of the century is not thematic but generic, and it occurs in two stages: at mid-century, within running prose, the voice of the White citation gives way to the name; after 1990 — in the American professional journal — the name moves into the note apparatus.

The prophetic word, then, did not disappear; it changed genre: the paper of the early century spoke with White to its readers, the paper of the late century refers to her.

2. Background

2.1 Three waves of White scholarship — and the missing fourth

The literature on White has been deposited in three clearly distinct waves, each asking a different question.

The first wave is denominational and biographical: the publications of the custodians of the estate, D. E. Robinson's history of the health message, Arthur White's multi-volume biography. Its question was what White said, and how her message fits the denomination's self-understanding.²

The second wave is critical-historical and begins with Ronald Numbers's Prophetess of Health (1976), which placed the origin of the visions and of the health message in contemporary American reform culture and thereby launched a decade-long controversy inside the denomination. Walter Rea's polemical source criticism (1982) and the officially commissioned Veltman report belong here as well. Its question: where does what White says come from.³

The third wave is consolidating: George Knight's hermeneutical and myth-dissolving works, Gilbert Valentine's administrative history, Ronald Graybill's textual history, and, as a summation, the Oxford volume Ellen Harmon White: American Prophet (2014), which treats the entire oeuvre outside a denominational frame. Its question: how the text came into being and how it is to be read — with literary assistants, revisions and compilations included.⁴

What is largely missing from all three waves is reception: how the White corpus lived in the everyday public sphere of the denomination. Knight asks a normative question — how she ought to be read — and his volume on the afterlife is anecdotal. The most recent apologetic collection points exactly here: in Anderson's characterization "Ellen White was a preacher and devotional writer," and "she was not even a systematic theologian"; Fortin reads her explicitly as a devotional writer; and Butler argues that for White herself and for her contemporaries the Testimonies were the "canon within the canon," while twentieth-century readers turned toward the devotional volumes.⁵ These claims, however, rest on case studies and reading experience, not on the run of the papers as a whole. Reception studies narrowed to a single paper and a short period do exist: a German thesis analyses the reception of White's writings in the 1895–1921 volumes of Zions-Wächter.⁶ What, to our knowledge, has not been done is a corpus-scale, multilingual, century-long quantitative measurement of reception.

This study attempts to open that fourth question: not what White wrote or where she took it from, but what her citations appear beside, in what textual form, and in what proportion, in the denominational press.

2.2 The theological stakes of citation form

In Adventism, citing White is not a matter of style but a doctrinal space. The denomination's fundamental beliefs see the gift of prophecy manifested in White's ministry and connect her writings to the Revelation phrase "the testimony of Jesus." At the same time the same tradition limits the status of the writings with White's own formulas: the "lesser light" leads to the greater, and — as a 1999 article in Curierul itself quotes — the testimonies were not given to take the place of the Bible.⁷ The placement of every single citation moves between these two poles: is it an argument from authority that decides, or a devotional accompaniment that illuminates? So when we measure whether a citation stands in an environment of doctrinal proof, life counsel, merchandise or footnote, we are measuring the history — embodied in editorial decisions — of the denomination's oldest internal debate: what the prophetic gift is for.

That debate unfolded at well-dated nodes: at the 1919 Bible Conference the leaders argued about verbal versus thought inspiration, and the minutes sank into the archive for half a century; Questions on Doctrine (1957) repositioned White's role for the sake of the evangelical dialogue; the source and inspiration crisis of the 1970s–80s made textual history public.⁸

This discourse can also be measured on our own corpus. Since the size of the corpus varies strongly by decade (75.9 million characters in the 1910s, 49.3 million in the 2010s), raw mention counts mislead; the table below therefore reports values normalized per million characters.

Table 1. The metadiscourse of White reception in the corpus (mentions per million characters)

Three observations follow.

First, the explosion of the source question in the 1980s exceeds every other marker by an order of magnitude (4.02 against 0.42 in the 1970s); in the 1990s the reception of the official answer — the Veltman report — takes its place, and in the 2000s a second, quieter source wave follows.

Second, the discourse of edition revision is not a product of the crisis. The value for the 1960s (0.20) is practically identical to that of the 1980s (0.19) — two decades before the plagiarism controversy. The institutional reason is known: in 1962 the White Estate issued a separate explanatory publication on the 1911 revision of The Great Controversy, and in 1963 Arthur White published a serialized account of the book's history in the Review and Herald.⁹ What is new here is not the existence of the campaign but its magnitude: the textual-historical speech of the denominational press already reached in the 1960s the intensity usually associated with the crisis of the 1980s.

Third, in the 2010s the strongest marker is the 1919 Bible Conference (0.67) — by then the denomination is no longer arguing about the textual question but historicizing its own history of controversy. Metadiscourse turns from crisis management into memory politics.

Two warnings came from the measurement itself. The phrase "the Bible and the Testimonies" in the first half of the century is not a boundary-defending claim but an established devotional collocation ("the study of the Bible and the Testimonies"), so an early "canon debate" computed from it would be an artefact. And the name Desmond Ford appears in the corpus from 1959 — as an author: the Avondale teacher was a regular Ministry contributor two decades before his name became, in 1980, the name of the crisis. The same journal printed his articles and later the controversy about him; the corpus preserves one of the most telling arcs of reception history as internal evidence.

2.3 Citation as an operation

The concepts for the question of citation form come from textual theory. Compagnon describes quotation as an operation of cutting and pasting: the quotation is the displacement of a foreign piece of text, which brings with it the voice of its source. Genette's typology of textual relations distinguishes degrees of explicitness of borrowing, from verbatim quotation to bare allusion. Hays's scale of biblical intertextuality — quotation, allusion, echo — applies the same gradation to religious textual practice.¹⁰ From these concepts a simple, measurable distinction follows: either the paper reproduces White's text (prints a quotation block) or it merely names her (gives a name, a title, a page number, without text). In the first case the authority speaks; in the second it is present but does not say anything. If the official press's use of White became institutionalized over time, this must have left a trace in the form as well.

2.4 Lifestyle norms: where authority is most visible

The Adventist normative system is most visible not in doctrinal propositions but in the rules of everyday life: eating, dress, amusement, reading, later cinema, radio and television. Two literatures in the study of religion explain why: boundary-work theory describes norms concerning the body and consumption as markers of communal boundaries, and strictness theory models the community-sustaining function of costly lifestyle norms.¹¹ The peculiarity of the Adventist case is that the authority anchor of this package of norms is the corpus of a single author — so norm and authority can be measured together.

Each normative domain has its own historiography. For diet, Numbers and the history of American health reform (the Graham–Jackson–Trall line) show that the Adventist health message is a denominationalized continuation of a broader reform culture.¹² The history of dress runs from nineteenth-century dress reform (Fischer) through the fundamentalist turn of the 1920s, which institutionalized the school ban on jewellery, to the loosening of the 1980s–90s — the General Conference permitted the wedding ring in 1986, while the Church Manual continued to recommend plain dress without ornament.¹³ Cultural consumption — theatre, novels, dancing, cards, then cinema, radio, television — is the most dynamic domain: from the prohibition list of the turn of the century to the late twentieth-century debate over "principle instead of rule."

How the weight of these domains shifted is shown by the corpus itself. The table below does not count White citations but the presence of topics in the entire American material (mentions per document, by decade) — that is, what the paper talked about at all.

Table 2. Presence of lifestyle and cultural-consumption topics (mentions per document)

Three movements can be read off. First, the vocabulary of prohibition fades — mentions of "amusements" as a collective category, of theatre, novels, dancing and cards fall to a fraction over the century; these are the elements of the classic nineteenth-century prohibition list. Second, media topics follow technology, not moral logic — radio is present from the 1940s, television from the 1950s, and both in a double register: as a moral danger ("abstaining from the radio" on the Sabbath; "television: blessing or curse?") and at the same time as an instrument of the church's own mission (radio evangelism, television fundraising). This duality is itself a product of the normative debate: the medium is first a danger, then a tool. Third, the weight of dress halves, while diet remains high throughout, peaking in the 1970s — and competitive sport is the only lifestyle topic whose presence grows.

These numbers measure the presence of the topic, not the strength of the normative edge; a report on radio evangelism and an admonition against the radio each count as one mention. It is precisely for this reason that normative operation is shown by measurement from the direction of the White citations, on which the rest of the study is built.

The turn from rule toward principle can also be read in the corpus. Richard Osborn's 1989 Ministry article on the cinema question states exactly this: the current emphasis creates the mistaken impression that where we watch something matters more than what we watch — the article draws on the literary criteria of George Knight's Myths in Adventism (1985) and argues that the distinction between the VCR, cable television and the cinema hall cannot be sustained.¹⁴ Another article in the same run of issues places the change of norms in a comparative frame: what Adventists can learn from Mennonites and other groups struggling with the same question.¹⁵ The debate, then, did not dispute White's authority but asked whether authority grounds a rule or a principle — and that distinction may leave a trace in citation practice as well.

The register of book distribution has a literature of its own: the history of religious print culture and colportage, which treats the religious book as both a mass medium and a commodity.¹⁶ The Adventist literature-evangelist institution is a late, highly organized form of it.

The normative and doctrinal layers of the Sabbath are treated by the standard literature of Sabbath theology;¹⁷ what matters for the present investigation is that the Sabbath is at once an identity background (vocabulary present everywhere) and a polemical doctrine (rare but with concrete markers) — and that, for measurement, the two come sharply apart.

2.5 Research questions

Six questions follow from the literature and from the findings above.

Q1. If 83.8–89.2% of citations do not stand beside the sanctuary complex, what registers do they stand in — and how large a share of them can be named at all?

Q2. Can the division of labour that the White literature assumes among the works (historical-prophetic vs. devotional vs. disciplinary corpus) be verified at the level of citations?

Q3. To what extent do lifestyle norms — eating, dress, cultural consumption, tithe, family education — operate as registers built on White's authority, and are they more stable than doctrinal themes?

Q4. Can the shift from rule toward principle be detected in citation practice: does the weight of normative lifestyle registers decline in the late twentieth century?

Q5. Does the form of the White citation change — from reproduced text to named, then referenced source — and if so, when and in which paper?

Q6. What can and cannot be measured with a dictionary in this material: which registers are lexicalizable, and what bias arises if this boundary is not made explicit?

3. Data and method

3.1 Corpus and basic concepts

The corpus, the citation detector and the theme coding are those of the first study: 7,133 issues, 177 million words, 78,091 accepted White citations, of which 74,520 fall in the nine analysed cells (three channels × three periods: 1920–1945, 1946–1989, 1990–2020). Distribution across the cells: Review and Herald 52,820, Ministry 17,936, Curierul Adventist 3,764 citations. "Theme hit" throughout means the presence of any of the eleven published themes in the ±50-word window; "residuum" means the set of windows with no theme hit.¹⁸ The residuum therefore means "not found," not "not there": the boundary rests on the published lexicon, whose manual validation is in progress.

3.2 Five steps to a map of the residuum

The analysis consists of five steps building on one another. The output of each step is a public file; the measurement steps were run with the seed and null density of the published run, so that the eleven theme rows of the first study reproduce bit-for-bit beside them (English branch: 273 shared cells, 0 differences; Romanian branch: 156 shared cells, 0 differences).

1. Work-level contrast. Citations are identified at work level, and the White corpus is thematically segmented at book level. As a first, lexicon-independent test we contrasted the source-work distribution of theme-hit and residual windows.

2. Discovery. The 32,163 residual windows of odd publication years were clustered with multilingual sentence embeddings (LaBSE, with per-track mean vector subtraction) and density-based clustering (UMAP + HDBSCAN, leaf selection). Even years were invisible to this step: they are the untouched half for later checking.¹⁹

3. Lexicalizability filter. From each cluster we distilled candidate expressions and measured whether the cluster's own expressions recover the cluster. Only a cluster that passed this test could become a dictionary theme.

4. Measurement. The five dimensions thus obtained were added to the project dictionary, and the whole machinery was re-run with the published seed — with lift, empirical null model and document-bootstrap intervals. The decisive test is the split-half: the expressions come from odd years, so a substantive lift proves something only on even years.

5. Centroid measurement. The clusters caught by the filter — the theological ones — were measured not with a dictionary but with embedding centroids: the centroid and the threshold were fixed on odd years, the measurement took place on even years. These numbers are shares without a null model; they cannot be placed beside the dictionary lifts as the same unit of measurement, and the text treats them separately throughout.

3.3 Measuring form

In the window of every citation we measure the length of the associated candidate quotation block: the word count of the quotation-marked span nearest to the citation, excluding the quotation marks of the work title. Three classes: no candidate block (the name or title stands without quotation), short (at most 35 words, i.e. one or two sentences), long (above 35 words). The measurement is an approximation: it does not see the unmarked block quotation (indented setting without quotation marks), and a broken OCR quotation-mark pair can distort it. Two checks close off the main artefacts: the normalizer converts typographic quotation marks into straight ones, so born-digital material (after 2005) is not a blind spot; and the direction of OCR loss works against the trend reported below — it would inflate the no-block share of the early period, not the late one — so the trend is, if anything, understated.

We also link form to the registers: for every citation we know whether its window carries the markers of the citation apparatus (step 4, section 5.3). Crossing the two signals separates naming in running prose from referencing in the note apparatus (section 6.3).

4. Composition: which work is cited when they are not teaching?

4.1 The division of labour among the works

The first result, independent of all machine theme coding, is the division of labour among the source works. We divided the share of citations standing in sanctuary-theme windows by their share in residual windows; a ratio above 1 indicates attraction to sanctuary language.

Table 3. Share of work families in sanctuary-theme versus theme-free windows (ratio, 95% document-bootstrap range)

The chi-square is overwhelming in all three channels (Ministry p≈10⁻⁸⁸, Review and Herald p≈10⁻¹⁵⁹, Curierul p≈10⁻⁹). The strongest single work is not The Great Controversy but Early Writings: its sanctuary attraction is 5.9-fold in Ministry, and 10.9-fold in the early period. When the investigative judgment has to be underpinned, the press reaches for the earliest, vision-adjacent text. The mirror image is equally sharp: the health works virtually disappear from the sanctuary windows.

4.2 The residuum as the terrain of the devotional canon

Table 3 sets the sanctuary themes against the residuum. Table 4 draws a wider boundary: it sets any theme hit (any of the eleven themes) against the residuum, broken down by individual works. The shares are computed within papers, and the cell is the ratio of the residual share to the theme-hit share; a value below 1.00 indicates a work pulling to the doctrinal side, above 1.00 one pulling to the residual side.

Table 4. Distribution of source works in residual versus theme-hit windows (within paper; residual% / hit% ratio)

Three findings follow.

The residuum is the terrain of the devotional book corpus. Steps to Christ, The Ministry of Healing, The Desire of Ages and Christ's Object Lessons are consistently overrepresented in theme-free windows — in all three channels, with two exceptions in the same direction. The doctrinal side calls on the visionary-historical set: White's periodical articles, Early Writings and The Great Controversy. The ratio for Early Writings is below 0.5 in all three papers: it is the most doctrinally bound work in the entire White repertoire.

Testimonies is proportional. Its ratio is 1.01–1.30: the work is present on both sides with roughly equal weight. Its dominance (26–40% of all citations, depending on channel) is pure volume, not a thematic profile — which is why normalization within papers was obligatory.

Seen within each work, the majority of every work is residual — but the ordering is steep: 64–89% of citations from White's periodical articles are theme-free, 73–74% of Early Writings, 78–84% of The Great Controversy, and 85–95% of every other major work. The two extremes are telling: 97.8% of Ministry's post-1990 citations from the health and dietary counsel compilations (Counsels on Health, Counsels on Diet and Foods) are residual, against only 52.4% of the same period's citations from White's Review and Herald articles.

The two tables together are an independent, two-sided confirmation of the work-level division of labour: prophetic themes call on the visionary works, while the carrier of everyday, non-doctrinal White presence is the devotional canon. This fits Butler's "canon within the canon" thesis precisely: the residuum is the terrain of those works toward which the twentieth-century reading public turned.⁵ The table is also the external anchor of the later steps: the source-work composition of the discovered clusters must reproduce the same pattern — and it does (5.1).

5. The registers of the residuum

5.1 A map of the residuum

The discovery run found 17 clusters; 37.0% of the residuum can be named, 63.0% remains diffuse. The nameable part falls into four qualitatively different registers.

Table 5. The registers of the residuum (odd years, discovery half)

Three observations belong here.

First, the larger half of the nameable residuum does not stand in an argumentative register — trade, apparatus and metadiscourse together make 19.8 percentage points against the argumentative 17.2. In the papers the White text also appears in the environment of merchandise, footnote and curricular accessory, not only beside arguments from authority.

Second, the source-work composition of the clusters independently reproduces the division of labour of section 4. The diet cluster is dominated by the health works (65%), dress and tithe by Testimonies (75% and 65% respectively), sanctification by the devotional works. Semantic embedding and title-string work identification are independent signals; their coincidence is the strongest internal argument for the taxonomy.

Third, metadiscourse — the press's own talk about the genesis of White's texts, the difference between the 1888 and 1911 editions of The Great Controversy, the literary assistants, later the Veltman research — is not a product of the plagiarism controversy of the 1980s. The cluster is continuously present from 1935 to 2015, on the Romanian track as well. The White text was an edited, annotated text long before it became a disputed text.

The 63% diffuse remainder is a single connected mass of density in semantic space, which clustering does not subdivide further without loss, and its distinguishing vocabulary is the most generic devotional vocabulary possible. This is itself a finding: the bulk of the residuum is not a hidden thematic order but a homogeneous stream of devotional speech into which the White citation is woven. It is at the same time the most important principled limit of the measurement: what is present everywhere cannot be seen by contrast-based measurement. This is also why the Sabbath — the constitutive word of denominational identity, evenly present in 6–9% of windows — could appear in the discovery neither as a cluster nor as a lexical theme; we return to it in 5.5 with a separately, top-down constructed dimension.

5.2 How far does the dictionary reach? The lexicalizability filter

The filter inserted between taxonomy and measurement is itself a result. From each cluster we distilled candidate expressions under document-frequency and precision conditions, and then measured whether the union of the candidates recovers the cluster.

Table 6. Does the cluster's own dictionary recover it? (English branch, discovery half)

The pattern is one of principle: concrete and administrative registers are lexicalizable, theological ones are not. Dietary counsel speaks in concrete nouns; sanctification is built from the same everyday devotional vocabulary as all other devotional prose, so even 24 expressions find only 7.9% of it. If we turned the whole taxonomy into dictionaries, the result would underrepresent precisely the largest and theologically most important categories — while measuring book advertisements accurately.

This table is at the same time a calibration of the first study's methodology. The terms of the sanctuary complex (investigative judgment, heavenly sanctuary, 2300 days, 1844) are of the well-lexicalizable, concrete kind; there the dictionary method worked within its reach. The present study shows where that reach ends.

5.3 The five measured registers

The five dimensions that passed the filter were run as a new block of the project dictionary, with the published seed and null density. They do not affect the theme-hit/residuum boundary: a window carrying only a new dimension remains residual in the sense of the first study and at the same time receives a register label — which is precisely the claim of this study.

Table 7. Lift of the five registers (±50 words, full corpus; cells distinguishable from chance in bold)

The split-half confirmed the main claims without shrinkage: the dietary dimension remains significant in all six cells on the held-out half as well (1.72–2.95); the late lift of the reference list "shrinks" from 12.22 to 11.78 and from 14.07 to 13.57; the early lesson schedule stands at 6.90. The original themes, run as a control, behave identically in the two halves, so the split itself is not confounded. The trade dimension is the weakest — on the held-out half it proves itself only in the middle period of the Review and Herald — and the text treats it accordingly.

The reference list. The trend of this dimension is the sharpest time series in the study: in Ministry the share of White citations standing in a note apparatus is 1.9% (1920–1945) → 15.3% (1946–1989) → 55.8% (1990–2020). The late value is distributed across 317 different issues, so it is not the artefact of a few bibliography-heavy numbers. The level is partly tautological — the detector looks for title mentions, and a bibliography is by definition a list of title mentions — so the result is not the level but the trend. The tautological component is the same in every period; the thirtyfold growth does not come from it. Section 6 approaches the phenomenon from the side of the form measurement as well.

The body as the most constant normative terrain. Dietary counsel is the only dimension significant in every channel and every period. White's authority is most continuous in the governance of the everyday body — more stably than in the core doctrinal themes, whose lifts fluctuate between 0.9 and 2.5. This fits well with the literature describing dietary norms as a boundary-marking practice of religious communities;¹¹ the Adventist peculiarity is that the authority anchor of the boundary marking is the corpus of a single author.

The disappearance of a genre. The Sabbath School lesson schedule produces one of the highest lifts in the whole investigation in the 1920–1945 band of the Review and Herald (8.86), then falls two periods later to 0.48. The explanation is not the death of the genre but its secession: the lessons were organized into a separate publication. This observation opens the planned third study, which examines the lesson quarterly as a corpus in its own right.

5.4 The non-lexicalizable registers: centroid figures

The theological clusters were measured with embedding centroids; the centroid and the assignment threshold were fixed on odd years, the measurement took place on even years. These are shares without a null model — they cannot be placed beside the lifts of Table 7 — but they answer two questions precisely: how large a place each register occupies in the residuum, and whether that place is stable.

Table 8. Share of the large non-lexicalizable registers in the residuum (even years, held-out half)

For all 17 registers the even-year share follows the odd-year share to within about one percentage point: the taxonomy is a property of the corpus, not of the discovery half.

Two substantive lessons. The sanctification register — the largest substantive block of the residuum — peaks at mid-century (17.6%), in the period of the denomination's internal devotional controversies. And the entire argumentative field shrinks after 1990, while the apparatus and meta registers grow: the composition of the residuum shifts from teaching toward apparatus. The same change of genre that the lifts of Table 7 measured is reproduced here by a completely independent instrument.

5.5 The Sabbath: what discovery could not see

It is striking that the Sabbath — the doctrine that gives Adventism its name — appeared nowhere in the taxonomy. The absence has a measurable cause, and it divides in two. Bare Sabbath vocabulary is evenly present in 6–9% of windows, in sanctuary-theme, other-theme and residual windows alike: a contrastless identity background that neither cluster nor lift can capture. Doctrinal Sabbath discourse (Sunday law, the mark of the beast, the seal of God), by contrast, is rare — 0.87% of residual windows — and was scattered in the clustering.

The doctrinal markers, however, are concrete and well lexicalizable, so they can be measured top-down; and since they derive from an a priori theological repertoire rather than from this corpus, the split-half requirement does not apply to them. The result:

Table 9. Lift of the Sabbath doctrine dimension

In the membership paper the White citation pulls significantly toward Sabbath doctrine in all three periods, and the binding strengthens — one of the few themes where the role of White's authority grows over time. In the ministerial journal there is no signal: the defence of the Sabbath is a genre of the lay public. The doctrinal Sabbath markers are also 2.7 times more frequent in sanctuary-theme windows than in the residuum: Sabbath eschatology moves in one package with the sanctuary complex, as the structure of The Great Controversy would lead one to expect.

5.6 The Romanian track: what can and cannot be measured

On the Romanian branch, the limits of measurability are themselves the results. The reference-list dimension works in Romanian too, and shows the same academicization (lift 2.17 → 3.40 → 7.22; 7.87 on the held-out half). The lesson schedule is significant in the early period (1.60), distribution in the late one (1.75). The dietary dimension cannot be measured in Romanian: a manual audit showed that nearly half the hits of the high-coverage Romanian expression (hrana) are book titles, biblical metaphors or scripture verses, so the dimension was restricted to clean expressions, beside which the number of cases (8–18 citations) no longer carries a claim. The difference between "no signal" and "not measurable" is not rhetorical here: the former would be a claim, the latter is a limit of the instrument.

Doctrinal Sabbath language practically does not exist in the environment of Romanian White citations (10 hits in a hundred years). This is consistent with the mechanism uncovered in the first study: between 1950 and 1980 Curierul published sensitive doctrinal content not as named White citation but with substitute attribution — "the spirit of prophecy says," "the servant of the Lord writes" — which a title-based detector cannot see.²⁰

6. Form: from voice to name, from name to footnote

6.1 Brevity is a house style, not a ritual mark

The initial hypothesis was that a substantial part of the residuum consists of short, motto- and epigraph-like, ritual citation. Half the hypothesis was confirmed: the short (at most 35-word) candidate quotation is indeed the largest form class of the residuum (48–54%). The other half was refuted, and that is the more interesting part.

Table 10. Form of citation in residual and theme-hit windows (nine cells together)

Form does not distinguish the residuum from doctrinal use: within each paper the distributions are identical to within a few percentage points, and the median length of the candidate quotation block is 11 words on both sides. Doctrinal witness works with text fragments just as short as devotional weaving. Brevity, then, is not a mark of ritual use but the house citation style of the denominational press.

The remainder of the motto hypothesis can be delimited with the register markers: 20.6% of short residual windows carry an apparatus, trade or lesson-schedule marker; the remainder — 23,247 "clean short" windows on the English branch — is the candidate motto set. Its work profile, however, is identical to that of the residuum as a whole: there is no separate motto corpus. If the motto register can be captured at all, it will be betrayed not by work and not by length but by position in the text (article-initial, article-closing placement); that test is still outstanding.

6.2 The formal movement of the century: the disappearing quotation block

The distribution of form across periods, by contrast, produces one of the strongest results of the investigation.

Table 11. Form of residual windows by period (no candidate block / short / long, %)

Between the two wars three quarters of invocations carry an actual quotation block. By the end of the century the ratio reverses: the majority of invocations carry no quotation block — they name White or her work but do not reproduce her text. The long quotation practically disappears (in Ministry from 7.0% to 1.4%). The pattern runs in the same direction and with the same magnitude in all three channels, runs the same way in theme-hit windows, and the two main sources of artefact are checked (3.3).

The shift can be named most precisely with a pair of concepts: over the century the White invocation moves from voice toward name. The early paper makes White's voice heard — it reproduces her sentences, often at length. The late paper calls her name: with work title and page number, increasingly without quotation. In the terms of 2.3: the place of quotation as cutting-and-pasting is taken by bare reference; the authority is present but no longer speaks.

6.3 Two stages: from running prose to the note apparatus

Table 11 presents as a single movement what is in fact two. If we cross form with the marker of the citation apparatus (5.3), the no-block invocations divide in two: some name White within running prose ("as Ellen White writes in The Desire of Ages…"), others stand in a note apparatus (reference list, numbered note, source statement at the end of the article). Table 12 reports this breakdown, now for all citations, not only for the residuum.

Table 12. No-block invocation and note apparatus, all citations (%)

The breakdown reveals two stages, separated in time and in geography.

First stage: from voice to name (mid-century, all three papers). Within running prose — that is, with the apparatus filtered out — the share of no-block invocation doubles between 1920–1945 and 1946–1989 (Review and Herald 24.3 → 51.6%, Ministry 25.0 → 50.7%), and begins the same way, on smaller numbers, in Curierul (16.3 → 35.4%). This change is not the spread of apparatus: it happens in running prose, in the middle of argument and exhortation. The paper no longer quotes from White but refers to her — in mid-sentence.

Second stage: from name to footnote (after 1990, the American professional journal). After 1990 the share within running prose barely moves (53.7%, 55.2%, 54.5%); what changes is the place of the reference. In Ministry more than half of citations (55.8%) stand in a note apparatus, in the Review and Herald nearly a quarter (22.8%), in Curierul only 7.1%. The second stage, then, is not universal: it is a product of the academicization of the ministerial professional journal, which touches the membership paper only partly and the Romanian paper hardly at all.

The two stages together give the content of this study's title. Over the century the prophetic word changed genre twice: first the reproduced text became an authority named in running prose, then — completely in one paper, partly in another — the named authority became a source referenced in the note apparatus.

This result is to be read together with the two-phase volume–binding movement of the first study. There it emerged that after 1980 citation becomes rarer and doctrinally specialized, and that in the 2010s volume is partly rebuilt while binding loosens. The formal result adds that the late rebuilding is already predominantly name invocation, and in the ministerial journal apparatus: White's presence is restored, but in a different state of matter — as reference, not as voice.

7. Discussion

7.1 Authority changes form, not scope

The secularization literature speaks of the narrowing scope of religious authority.²¹ Our data permit a finer claim: in the official press White's authority does not disappear, nor does it retreat into a narrower range of topics — according to the first study its doctrinal binding in the American papers even strengthens for a while after 1980. What changes is the form of the authority. In the late ministerial journal the White sentence does not sound from the pulpit; it certifies from the footnote.

7.2 The measurable shape of routinization

The three cross-sections — composition, register, form — point in one direction, and together give the formal imprint of the routinization of charismatic authority.²² Composition shows that the carrier of everyday, non-doctrinal White presence is the devotional canon: precisely those works toward which, in Butler's account, the modern Adventist reading public turned. The registers show that the larger half of the nameable residuum is not argument at all but the institutional life of the White corpus: its distribution, its apparatus, its self-reflection. And form shows that the physical gesture of invoking authority is itself institutionalized: the reproduction of the text is replaced by the display of name and title.

The classic description of routinization speaks of the depersonalization of charisma: the founder's authority is rebuilt into office, textual corpus, procedure. The present data show a step of this that has not been measured before: the textual corpus itself becomes depersonalized. As long as the paper prints White's sentences, the authority is a speaking authority — it has a voice, a rhythm, a rhetoric. When the paper merely names her ("see The Great Controversy, p. 488"), the authority becomes a directory entry: its presence is assured, its content optional. The canonical status of the White corpus and the 85% residuum can thus both be true at once: canonical status shows itself precisely in the fact that one no longer needs to quote — it is enough to point.

7.3 Rule and principle

The answer to Q3 is unambiguous: among lifestyle norms, diet is the most stable register built on White's authority, and in this it surpasses the core doctrinal themes. The answer to Q4 is partial. The vocabulary of prohibition fades from the papers' subject matter (Table 2); according to the centroid measurement the argumentative registers — among them administrative and family counsel — shrink after 1990 (Table 8); and the corpus itself documents the debate turning from rule toward principle (Osborn, Rydzewski). The dietary register, however, does not weaken: the governance of the body is the area where White's authority still grounds a rule at the end of the century. Whether on the other normative domains authority turned from rule into principle, or whether it is simply cited less there, the register measurement does not decide; that is a question for the function coding.

7.4 A methodological lesson

The result of the lexicalizability filter — that theological registers cannot be captured with dictionaries — is a general lesson for all dictionary-based research on religion: contrast-based measurement is systematically biased toward concrete registers, and this bias can only be stated, not removed. The present study places two instruments — dictionary and centroid — side by side, and where they measure the same movement (the advance of the apparatus), they agree.

7.5 The alternative explanation

One more prosaic explanation must be excluded, and we can do so only in part: the general change in citation conventions in twentieth-century publishing (shorter articles, editorial style sheets, the spread of footnote culture) could by itself have reduced the share of block quotations. Separating the two stages answers this in part: the second stage is channel-dependent (55.8% in Ministry, 7.1% in Curierul), so it is not a universal press convention but the academicization of one paper. The first stage, by contrast, is identical in all three papers, and is compatible with a general change of convention. To separate it out, the formal trajectory of non-White quotations (biblical and contemporary authors) would have to be measured in the same way in the same corpus. This is feasible, and is among the steps to be taken before publication.

8. Limitations

The results measure thematic and generic co-occurrence, not rhetorical function; whether a given citation proves, illustrates or sells is to be decided by the manual function coding now under way. The theme-hit/residuum boundary rests on the lexicon of the eleven published themes, whose precision and recall have not yet been estimated on a manual sample. The positive characterization of the diffuse 63% is only indirect (source-work profile, generic vocabulary). The centroid figures are threshold-dependent, and for two diffuse commercial clusters the threshold is close to zero — there the dictionary measurement is authoritative. The form measurement is an approximation: it does not see the unmarked block quotation, and does not bind to the citation a quotation-marked span more than 250 characters away. The position-based test of the motto hypothesis has not yet been carried out. OCR noise reduces coverage in the early volumes; column bleed can sweep foreign text fragments into the windows. The Romanian case numbers in the middle period are at or below the limit of measurability for every dimension. The corpus is limited to the printed official organs.

9. Conclusion

The doctrine-free majority of White invocation is not a shapeless remainder but a structured space: in its source works the devotional canon; in its registers devotional prose, lifestyle counsel and the institutional life of the corpus — its distribution, its apparatus, its self-reflection; and in its form a century-long, two-stage shift from reproducing the text to displaying the name, and from the name to the footnote. The most tenacious terrain of authority use is not doctrine but the body: dietary counsel is the only register that, across a hundred years, attracts the White citation in every paper and every period.

The White of the official paper speaks at the beginning of the century; at the end of the century she is referenced. If the first study measured what the authority was called upon for, this study measures the state of matter in which it is present — and the answer points the same way: the prophetic word did not disappear, it changed genre.

Data, code and reproducibility

Same corpus, same citation detector, same ±50-word window, same empirical null model as the first study (seed 20260814, 100 null windows per 10,000 words, 400 document-bootstrap replicates). Null density is not a negligible setting: the default density of 20 shifts every lift with the same observed side; the figures reported here were produced with a density of 100. The extended run computes the eleven published themes unchanged: in the 273 shared cells of the English matrix and the 156 of the Romanian, the difference is zero.

The figures of every table derive from project files: metadiscourse and lifestyle topics (background_mine.py, background_{meta,lifestyle}_table.txt), work-family contrast (egw_work_theme_contrast*.csv), per-work crosstab (work_residual_kereszttabla.csv), register summary (cluster_registers_report.txt), lexicalizability (lexicalizability_report.txt), R-series lifts and split-half (theme_{en,ro}_r_matrix.csv, splithalf_{en,ro}.txt), centroid measurement (centroid_measurement.csv), Sabbath dimension (theme_*_r_matrix.csv, S_SABB rows), form measurement (citation_quoteform_{en,ro}.csv, row-aligned with the theme coder's output), and the crossing of form with apparatus (form_register_link.py, form_register_link.txt). The curation rationale for the dictionary extension is documented in the comments of the R and S blocks of lexicon.yaml, including the discarded expressions.

Steps before publication. (1) Completion of the manual validation (theme precision sample in both languages, function coding with two coders, Curierul borrowings), and incorporation of the results. (2) Manual spot-check of the precision of the five register dictionaries on the English branch, on the model of the Romanian audit. (3) The position-based test of the motto hypothesis. (4) Control trend: the formal trajectory of non-White quotations in the same corpus (7.5).

Notes

Appendices

A. Parameters of the discovery run

LaBSE sentence embedding with per-track mean vector subtraction (without it the clusters separate by language: 0/15 mixed clusters; after centring 14/15, Romanian–English anchor retrieval rises from 0.20 to 0.55, anchor purity changes from 0.75 to 0.72). Model selection on a three-way test (LaBSE / paraphrase-multilingual-mpnet / multilingual-e5): e5 has the best monolingual resolution but zero cross-lingual retrieval — unsuitable for a bilingual design. UMAP to 5 dimensions, HDBSCAN min_cluster_size=150, leaf selection; excess-of-mass selection would have swallowed about 88% of the residuum into a single cluster — residual prose is a single connected mass of density in semantic space, and the study treats this methodological fact as a claim.

B. Known error classes and their handling

Check on the suspicion of apparatus: the share of apparatus markers is 29.3% in the residuum and 27.1% in sanctuary-theme windows — apparatus is not peculiar to the residuum, so it does not distort the first study's proportions; the narrowly commercial register, by contrast, is enriched in the residuum (price/advertisement ×2, colporteur ×5) and is to be treated as a separate category. Detector overrun (the 7 issues downloaded after the reference set) is marked in_ref=0 in the export. The Romanian expression hrana was demoted after a manual audit (5.6). Two metadiscourse markers were withdrawn after inspection and are reported as findings (2.2): the early collocation "the Bible and the Testimonies," and the occurrences of Desmond Ford's name before 1970, which are authorial bylines.

C. Manual validation plan

The obligatory round before publication is the same as the first study's, and partly shared with it: (a) a theme-stratified precision sample in two languages — the Romanian windows with translation-mediated coding, which the validation appendix states explicitly; (b) function coding of 302 windows with two coders (proving / illustrating / disputing / other), with inter-coder agreement; (c) an English spot-check of the five new dimensions. The validation package is in prepared state in the validation/ directory.