Disputed authorship (textual analysis)

Who wrote a disputed text, and how much of it? Bayesian evidence can weigh hypotheses in which a text has more than one author; here it is applied to the Federalist Papers, Shakespeare's collaborations and books of the Bible.

Schematic in two parts. Left: a page of text drawn as rows of short rounded dashes on a pale card; most dashes are light gray (content words, not counted) and the shorter function-word dashes are tinted violet (first hand) or orange (second hand), violet outnumbering orange roughly two to one; a key below reads first hand, second hand, content words not counted. Right: a horizontal axis labeled share of the second hand, running from 0 through one half to 1, with a smooth violet curve labeled posterior that rises from zero, peaks near 0.2 and falls away before a dashed vertical line at one half labeled even split.

A disputed passage with its function words colored by author. The test models the rates of these words as a blend of two authors’ profiles and, from their counts alone, computes a probability distribution for the second author’s share. Content words (gray) are not used. (Illustration only.)

Deciding authorship by counting

Who wrote a text is ordinarily settled by evidence outside it: a title page, a printer’s register, a letter. When that evidence is missing or contradictory, scholars turn to the language of the text itself. Attempts to decide such arguments by counting began in the late nineteenth century. T. C. Mendenhall compared authors by curves of word length in 1887, the word “stylometry” comes from Wincenty Lutosławski’s 1897 book on Plato, and G. Udny Yule studied sentence lengths in 1939. Neither Mendenhall nor Yule attached a probability to a hypothesis.

Frederick Mosteller and David Wallace brought probability into the question in 1963 and 1964, working on the Federalist essays of 1787–88. Hamilton left a list of his essays the day before his duel with Burr; Madison’s appeared in 1818. The lists disagree on twelve essays, three more (Nos. 18, 19 and 20) are acknowledged joint work, and five are John Jay’s. The two men proved “practically twins” in sentence length, so Mosteller and Wallace separated them by the rates of function wordsClosed-class words such as articles, prepositions, conjunctions, pronouns and particles. Their rates vary more from writer to writer than from subject to subject, which is why authorship tests count them and leave content words out.: upon occurs at three per thousand words in Hamilton and one sixth per thousand in Madison. They computed posterior oddsThe odds between two hypotheses after the data are taken into account: the prior odds multiplied by how much better one hypothesis predicts the observed counts. for each disputed essay and gave all twelve to Madison; Nos. 18 and 19 they judged mainly his.

Most later work treats attribution as classification, assigning each text to a candidate author; where a text has more than one hand, the methods in use locate where the style changes or divide the text among a given number of authors. The composite hypothesis itself, that an essay is one-third Hamilton’s or that a book had four sources, is seldom given a probability. Here those probabilities are computed for the Federalist, for Shakespeare’s plays and for the books of the Bible; most of the answers agree with the scholars, and a few do not.

Texts as mixtures

A text is reduced to the counts $U = (U_1, \ldots, U_k)$ of $k$ listed function words, treated as independent draws. A hypothesis about who wrote it constrains the probability $p_v$ that a token (any one running word of the text) is word $v$, and the hypothesis is weighed by its evidence: the probability of the observed counts once every unknown rate, share or boundary has been integrated over its prior. For one homogeneous text with a symmetric Dirichlet priorA standard prior over a set of proportions that must sum to one; symmetric means every listed word is treated alike before any text is seen. of parameter $\alpha$ on its rates, the evidence is

$$Z_\alpha(U) \;=\; \int \prod_{v=1}^{k} p_v^{\,U_v}\; \mathrm{Dir}(p \mid \alpha)\, dp \;=\; \frac{\Gamma(k\alpha)}{\Gamma(k\alpha+N)} \prod_{v=1}^{k} \frac{\Gamma(\alpha+U_v)}{\Gamma(\alpha)}\,, \qquad N = \sum_{v=1}^{k} U_v\,.$$

Here $N$ is the total count of list-word tokens in the text, $\Gamma$ is the gamma function (the factorial extended beyond whole numbers), and $\mathrm{Dir}(p \mid \alpha)$ is the prior on the rate profile $p = (p_1, \ldots, p_k)$, which treats all $k$ words alike; the source-counting analyses below are run at three settings of $\alpha$. All the evidence integrals are evaluated in closed formAn explicit formula that can be evaluated exactly in a finite number of standard operations, with no simulation and no numerical approximation., many in exact rational arithmetic, with no sampling error. For a text that two known authors may share, the profiles $p_A$ and $p_B$ are the word rates of each author’s undisputed writings, held fixed. Under the blend hypothesis each token comes from $A$’s profile with probability $w$, an unknown share with a uniform prior between 0 and 1:

$$p_v(w) \;=\; w\, p_{A,v} + (1-w)\, p_{B,v}\,, \qquad Z_{\mathrm{blend}}(U) \;=\; \int_0^1 \prod_{v=1}^{k} \bigl[\, w\, p_{A,v} + (1-w)\, p_{B,v} \bigr]^{U_v}\, dw\,.$$

The pure hypotheses are the endpoints $w = 1$ and $w = 0$. The integrand is a polynomial in $w$, so the integral is exact, and the same polynomial, normalized, is the posterior density of $w$; for the Federalist $A$ is Hamilton and the posterior mean of $w$ is the Hamilton share. In real texts, though, an author’s rate for a word varies from text to text more than independent draws allow, as Mosteller and Wallace found. Every two-author analysis is therefore repeated under a second, hierarchical modelA model with two levels of chance: each text’s word rates scatter around its author’s profile, and the words are then drawn from that text’s own rates. in which each text’s rates scatter about its author’s profile by an amount fitted to that author’s undisputed texts; its evidences are again finite sums in closed form.

For a book whose sources may lie end to end, a hypothesis divides its chapters into $g$ contiguous segments, each with its own rates integrated out by the first formula. Summing over every placement of the boundaries gives a posterior over $g$ and over where the cuts fall. The source of each segment is called a “contiguous generating process,” which need not be an author: a seam can mark a change of genre or language as easily as a change of hand. Two hypotheses are compared by their Bayes factor,

$$\log_{10} \mathrm{BF}(H_1 : H_2) \;=\; \log_{10} \frac{Z(U \mid H_1)}{Z(U \mid H_2)}\,,$$

positive values favoring $H_1$, each unit a factor of ten in the odds. Two cautions apply throughout. First, the magnitudes depend on the count model, so conclusions are drawn only from the side that the posterior of $w$ favors, leaving aside where it peaks and how wide it is, and, for the books, from the boundary positions that stay fixed across prior settings. Second, a negative result carries weight only where the detection power, the chance of catching a division that is really there, has been measured, here by inserting a second author’s text into single-author works. A play qualifies only if a fifth of its verse, replaced by that of any comparison dramatist of its period, would change the classification at least eight times in ten; a biblical book needs roughly 4,800 function-word tokens.

The Federalist and Hamilton's share

The Federalist analysis uses three word lists, each of which assigns 64 of the 65 papers of known authorship correctly when a paper is left out and scored: the seventy function words Mosteller and Wallace screened, the 318-word English stop-word listA list of very common words (the, of, and, to, that and so on) that text-processing software sets aside; nearly all of them are function words. of the scikit-learn software library, and the 300 most frequent words. Under all three the joint papers are Madison-dominant (first figure). The posterior mean Hamilton share is at most 0.071 for No. 18, 0.230 for No. 19 and 0.258 for No. 20. Papers Madison wrote alone, left out of his profile in turn, receive comparable or larger shares (medians 0.13, 0.19 and 0.26 by list), so no Hamilton contribution above that level is resolved in the joint papers. Under the second model both sets of numbers fall (the largest joint-paper mean to 0.094) and the comparison is unchanged.

Dot-and-bar chart. Horizontal axis: Federalist paper, with Nos. 18, 19 and 20 under the heading acknowledged joint papers, then, after a thin gray divider, Nos. 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 62 and 63 under the heading the twelve disputed papers. Vertical axis: Hamilton share w from 0 to 0.9. For each paper three marks stand side by side: slate for the Mosteller–Wallace 70-word list, violet for the 318 stop words, orange for the 300 most frequent words; each dot is a posterior mean and each vertical bar runs from the 5th to the 95th percentile. A dashed horizontal line marks the even split at w = 0.5. Three dotted horizontal lines in the list colors, at 0.13, 0.19 and 0.26, mark the median share the same analysis gives a paper Madison wrote alone. All nine joint-paper dots lie at or below 0.26, No. 18 lowest near 0.04 to 0.07 and No. 20 highest near 0.12 to 0.26. Among the disputed papers only No. 55 under the slate list has a dot above 0.5, at 0.51 with a bar from 0.28 to 0.75; its violet and orange dots sit at 0.32 and 0.39. Every other disputed-paper dot lies below 0.35, most between 0.05 and 0.3, with bars a few tenths wide. A legend below the axis names the three lists and the two kinds of reference line.

Hamilton’s share $w$ of each contested Federalist paper under the three word lists. A dot is the average share the word counts point to (the posterior mean) and a bar covers the middle ninety percent of the posterior. The dashed line is an even split. The dotted lines show, one per list, the share the same calculation assigns to a typical paper that Madison wrote alone, so a dot near or below them cannot be told from pure Madison. All three joint papers sit at or below that level, and No. 55 is the only disputed paper whose mean passes one half, under one list only. (Drawn from Tables 3 and 4 of the paper.)

Of the twelve disputed papers only No. 55 has a Hamilton share above one half under any list, and only under the inherited seventy words (0.5146). Under the other two lists, which did better on the known papers (the same number of correct calls, but stronger evidence for the true author on average), its shares are 0.3198 and 0.3879 and the pure-author comparison favors Madison: a Madison majority with a Hamilton share near a third. That share may reflect a Hamilton contribution or paper-to-paper variation that the independent-draws model leaves out, and these data cannot tell the two apart. The second model, which builds that variation in, lowers the shares to 0.075 and 0.132, but it also misses a constructed even mixture of two papers about a third of the time, so the drop does not settle whether No. 55 is joint work either.

Shakespeare's collaborators

Doubts about sole authorship in the Shakespeare canon go back at least to Edmond Malone in 1787, and in 2016 the New Oxford Shakespeare printed Marlowe’s name beside Shakespeare’s on the Henry VI title pages. Here all 41 plays are scored the same way, on 365 function words. For each play, a blend of Shakespeare’s profile with that of one other dramatist (eight candidates, matched by period) is compared with the better of the two single profiles. Securely solo plays of the same period, scored the same way, also show some gain from blending, and the largest such gain sets a ceiling. A play is classed as having a second hand only if its gain exceeds that ceiling by a set margin and at least one act is decisive on its own.

Six plays pass this test: 2 and 3 Henry VI, Edward III, Henry VIII, The Two Noble Kinsmen and Titus Andronicus. Four show no second hand, and for these the test demonstrably had the power to find one: when a fifth of the verse is replaced by verse of any comparison dramatist of the period, the classification changes at least 84 times in a hundred. They are King John, A Midsummer Night’s Dream, Richard III and The Tempest. For eight of these ten plays the result agrees with the published attribution. For 2 and 3 Henry VI, where scholars are divided between revision and collaboration, a second hand is detected in both.

The other thirty-one plays cannot be resolved, twenty-four of them because the test lacks power at their length or against their comparison set. In Hamlet and seven others a planted fifth changes the classification only 9 to 52 times in a hundred. For sixteen more a simpler estimate of the power already falls short or is marginal; nine of these are late plays, Macbeth and Pericles among them, for which solo plays of the period already score high against Middleton’s profile. In the remaining seven the gain clears the ceiling but fits no single pair of profiles. The shares scholars have proposed for these unresolved plays are not tested here, and the results say nothing against them.

On two finer points the results depart from the scholarly attributions. In Henry VIII the second hand, given to Fletcher since 1850, is closer on function words to Middleton’s profile (log Bayes factor 62.55) than to Fletcher’s, which barely clears its own ceiling (3.34 against 2.36). Under the second model only Middleton’s gain stays decisive. Two dramatists writing in the same late idiom may each fit the other’s profile, so this need not mean a different collaborator. In Titus Andronicus Peele’s profile is decisive in act 3, whereas Brian Vickers’s division gives him chiefly act 1; Marlowe’s profile also clears its ceiling and is decisive in act 1, and under the second model the classification rests on that act alone.

The three parts of Henry VI are then analyzed more closely, against profiles of Shakespeare, Marlowe and Kyd, on a 142-word list chosen by testing nine reference plays. In all three a Shakespeare–Marlowe blend beats the best single profile, by $10^{9.9}$, $10^{12.7}$ and $10^{30.9}$, with Shakespeare shares of 0.70, 0.64 and 0.48; Kyd is never preferred. Left out of its own profile and scored the same way, Edward II, accepted as Marlowe’s unaided work, receives a blend factor of $10^{38.8}$ and a share of 0.49. A blend preference as strong as any part’s can therefore arise without a second hand, and by itself does not show collaboration. Under the second model only Part Three’s preference stands clear of what solo plays produce. Edward II is also the one reference play the 142-word list misclassifies, so every Marlowe-side number here is conditional on that one play.

Two-part graphic. Left, under the heading class of each act, a grid of fifteen rounded tiles in three rows labeled 1 Henry VI, 2 Henry VI and 3 Henry VI and five columns labeled act 1 to act 5. Each tile is violet with a white S (Shakespeare-side), orange with a white M (Marlowe-side) or pale gray with a question mark (uncertain). Row 1 Henry VI: S, S, S, uncertain, S. Row 2 Henry VI: uncertain, S, S, M, S. Row 3 Henry VI: M, S, uncertain, M, uncertain. A key below repeats the three tile types, and a footer reads verse only, 142 function words, profiles of Shakespeare and Marlowe. Right, under the heading Shakespeare share w of the whole play, a horizontal scale from 0, labeled all Marlowe, to 1, labeled all Shakespeare, with a dashed line at one half. Level with each row a slate dot with a bar: 0.70 with bar 0.62 to 0.77 for Part One, 0.64 with bar 0.57 to 0.72 for Part Two, 0.48 with bar 0.41 to 0.55 for Part Three. Below them an open orange diamond at 0.49 labeled for comparison: Edward II, by Marlowe alone, left out of his own profile. Footer: dot, posterior mean; bar, 5th to 95th percentile.

The three parts of Henry VI compared with profiles of Shakespeare and Marlowe. Left: the verse of each act is classed Shakespeare-side (S), Marlowe-side (M) or uncertain (?) according to whether its evidence lies beyond nearly every value that reference scenes of the other dramatist produce at that length. Marlowe-side acts appear only in Part Three, acts 1 and 4, and in act 4 of Part Two, the Jack Cade rebellion. Right: the Shakespeare share $w$ of each whole play falls across the trilogy from 0.70 to 0.48. The diamond is a caution: Edward II, which Marlowe wrote alone, receives 0.49 when scored the same way, so a share near one half does not by itself show a second hand. (Act classes and shares from Figure 2 and Table 6 of the paper.)

When each act is classed on its own (second figure), Marlowe-side acts occur only in Part Three, acts 1 and 4, and in act 4 of Part Two, the Jack Cade scenes that Hugh Craig and Arthur Kinney associated with Marlowe in 2009. The agreement with Craig and Kinney is conditional, however: the Part Two act clears its threshold by only 0.11, and without Edward II in Marlowe’s profile all three acts favor Shakespeare.

Isaiah and the sources of the Pentateuch

The division of Isaiah goes back to the eighteenth century, when Döderlein and Eichhorn assigned chapters 40 onward to a later author writing in the Babylonian exile; in 1892 Bernhard Duhm cut the book in three, chapters 1–39, 40–55 and 56–66. Here the number of sources is not fixed in advance: a posterior is computed over that number and over where the boundaries fall, from the 66 chapters of the Leningrad Codex of 1008 CE, the oldest complete manuscript of the Hebrew Bible.

Two horizontal strips of 66 small cells, one cell per chapter of Isaiah, with chapter numbers 1, 10, 20, 30, 40, 50, 60 and 66 beneath each strip; cells 1 to 39 are tinted pale slate and cells 40 to 66 pale violet. Above both strips a thin gray reference bar labeled Duhm’s division (1892) is split into three spans: First Isaiah, chapters 1 to 39; Second, 40 to 55; Third, 56 to 66. Upper strip, labeled Leningrad Codex, 1008 CE, 66 chapters: a solid violet bar between cells 39 and 40 labeled after ch. 39, posterior 0.98 to 0.99; a thinner, paler violet bar between cells 35 and 36 labeled after ch. 35, if a third segment is allowed; cells 36 to 39 tinted slightly darker with a note beneath, Hezekiah narrative (prose), chapters 36 to 39; an orange dashed line between cells 55 and 56 labeled after ch. 55, posterior 0.000. Lower strip, labeled Great Isaiah Scroll, copied about 125 BCE: a pale violet band covering cells 37 to 40 with a solid violet bar between 39 and 40, labeled within chapters 37 to 40: posterior 1.00, most probably after ch. 39 (0.57 to 0.68); a slate dashed line between cells 33 and 34 with the note change of scribe after ch. 33: posterior 0.000; an orange dashed line between 55 and 56 labeled after ch. 55, posterior 0.000. A key at the bottom reads: violet bar, boundary found by the function-word counts; orange dashed line, traditional Second/Third Isaiah seam (posterior 0.000); slate dashed line, change of scribe.

Where the function words divide Isaiah. Each strip is the book’s 66 chapters in one manuscript: the Leningrad Codex of 1008 CE and the Great Isaiah Scroll from Qumran, about a thousand years older. The gray bar on top is Bernhard Duhm’s 1892 division into three prophets. On both manuscripts the counts place a boundary at the end of chapter 39 (violet), where Duhm’s First Isaiah ends; if a third segment is allowed, the codex adds one at the edge of or inside the prose Hezekiah narrative of chapters 36–39. At Duhm’s second seam, after chapter 55, the boundary probability is 0.000 on both (orange), although in the short later chapters even a change as large as the one at chapter 40 would be caught only about six times in ten. The scroll’s change of scribe after chapter 33 (dashed) leaves no trace, as expected of a statistic that is blind to handwriting. (Numbers from Section 6.2 of the paper; its Figure 3 shows the codex posteriors in full.)

For Isaiah a single process is rejected at every prior setting, by log Bayes factors of 41.5 to 98.2. Given two processes the boundary falls after chapter 39 with posterior 0.979 to 0.994; given three, the extra boundary falls at or inside the prose Hezekiah narrative of chapters 36–39 (third figure). A boundary after chapter 55 has posterior 0.000, given three or four processes, at every chapter-level setting. That zero concerns function-word rates and sharp boundaries only: the case for a third Isaiah rests substantially on content, theology and historical setting, which the word list excludes by design, and gradual drift would not show as a sharp cut. The test is also weaker there, since chapters 40–66 are short: a change at 55|56 as large as the one measured between chapters 1–39 and 40–66 would be caught only about six times in ten, and a smaller one would usually be missed.

The Great Isaiah Scroll from Qumran, copied around 125 BCE and a millennium older than the codex, is a second witness. On the scroll a single process is again rejected, at 44.0 to 85.3. Given two processes the boundary lies within chapters 37–40 with posterior 1.0000 at every setting, most probably after chapter 39. At one setting the scroll’s posterior for the number of processes spreads over three to six, so the two manuscripts agree on the boundary but can differ on the count. The scroll also allows a check on whether the statistic responds to scribal hand: handwriting analysis indicates that two scribes probably copied one half each, with the seam after chapter 33, and the boundary posterior there is 0.000. After chapter 55 it is 0.000 on the scroll as well, so neither manuscript supports a third Isaiah, within the limited power that the short later chapters give the test.

The Pentateuch is a different case: on the documentary hypothesis, to which Julius Wellhausen gave its classical form in 1878, the five books interleave, verse by verse, four sources now called J, E, D and P. Interleaved sources cannot be contiguous segments, so the verse-by-verse assignments tabulated by Carpenter and Harford-Battersby in 1900 are scored as fixed hypotheses, by how well each source map predicts blocks of text left out of it. Separating the Priestly P verses improves prediction by about 1,600 nats (natural-log units; 2.303 nats make a factor of ten), 13.2 standard deviations above the mean of random relabelings of the blocks, and P stays distinct when the comparison is confined within single books or within narrative alone. Adding the Deuteronomic D gains a further 943 to 966 nats. Nearly all D material, however, is the book of Deuteronomy, and with labels shuffled only within books its gain is no longer exceptional, so on function words D coincides with the book itself. Splitting the rest into J and E does not help: the gain is $-18.1$ to $-27.2$ nats, inside what fake partitions of single-source Esther produce by chance (up to $+33.5$). Here the test did have power: a J/E difference two-thirds to five-sixths the size of P’s contrast with the rest would have been caught at least eight times in ten. The divine names, doublets and theology on which the J/E case rests are excluded by construction, so this null result does not refute the documentary hypothesis, just as the separation of P and D does not by itself prove that they were documents.

Every book of the Bible

The same boundary-finding model (a changepoint modelA model for a sequence whose statistical properties switch abruptly at unknown positions; the analysis infers how many switches there are and where they fall., in statistical terms) is then applied unchanged to every book of the Bible. Twenty-one are classed as composite, nine show no division at their length, and thirty-six, including twenty-one of the twenty-seven New Testament books, are too short for any result. Neither kind of non-result is evidence of unity; each reflects only the test’s power at that length. Many composite results fall where source criticsScholars who reconstruct the earlier documents and editorial layers from which a biblical book was compiled. put a division, as in Leviticus at the Holiness Code and Deuteronomy at its law code. For Ezekiel the result departs from the commentators, who usually read it as the most unified major prophet: it is the strongest single-language composite in the survey (log Bayes factor 363.5). It divides after chapter 39, where the prophecies give way to the temple vision, a seam that may mark the change of subject as much as any change of hand.

When whole books within each section of the canon are grouped by their word rates, the synoptic gospels (Matthew, Mark and Luke) share one process, as Matthew’s and Luke’s use of Mark predicts. Acts, however, separates from Luke, even though most commentators assign both to one author; the counts show only that the two books differ in their function-word rates, whoever wrote them. Comparing all sixty-six books at once shows that several profiles are shared across sections of the canon; at one prior setting, for example, the synoptic gospels group with James. Merged wherever the evidence favors a shared profile, the books fall into roughly a dozen to thirty distinguishable function-word profiles (13 to 32, depending on the prior and on the vocabulary compared). These figures count profiles at this resolution and say nothing about the number of authors or sources, even as a bound. When the analysis is repeated on the oldest witnesses, Codex Sinaiticus and the Septuagint among them, 19 of 26 checkable books keep their classification on their earliest witness, and 25 on at least one line of transmission.

The paper

Supplementary material

All texts analyzed are public editions.

References

T. C. Mendenhall, The characteristic curves of composition, Science ns-9 (1887) 237compared authors by curves of word length, with no probability attached to a hypothesis
W. Lutosławski, The Origin and Growth of Plato’s Logic, with an Account of Plato’s Style and of the Chronology of his Writings, Longmans, Green (1897)the book on Plato’s dialogues that coined the word “stylometry”
G. U. Yule, On sentence-length as a statistical characteristic of style in prose, with application to two cases of disputed authorship, Biometrika 30 (1939) 363sentence length as a statistical marker of authorship
F. Mosteller and D. L. Wallace, Inference in an authorship problem, J. Amer. Statist. Assoc. 58 (1963) 275posterior odds for the disputed Federalist essays from function-word rates
F. Mosteller and D. L. Wallace, Inference and Disputed Authorship: The Federalist, Addison–Wesley, Reading MA (1964)the full study: all twelve disputed essays to Madison; source of the seventy-word list
E. Malone, A Dissertation on the Three Parts of King Henry VI, London (1787)the canon’s first modern editor; argued that the three Henry VI plays were not originally Shakespeare’s
J. Spedding, Who wrote Shakspere’s Henry VIII?, Gentleman’s Magazine, n.s. 34 (1850) 115divided Henry VIII scene by scene and gave the second hand to Fletcher
B. Vickers, Shakespeare, Co-Author: A Historical Study of Five Collaborative Plays, Oxford University Press, Oxford (2002)the scholarly divisions of five collaborative plays, Peele’s act 1 of Titus Andronicus among them
H. Craig and A. F. Kinney (eds.), Shakespeare, Computers, and the Mystery of Authorship, Cambridge University Press, Cambridge (2009)computational studies of the canon; associated the Jack Cade scenes of 2 Henry VI with Marlowe
G. Taylor and G. Egan (eds.), The New Oxford Shakespeare: Authorship Companion, Oxford University Press, Oxford (2017)the current co-author attributions, with Marlowe’s hand in the Henry VI plays
J. C. Döderlein, Esaias ex recensione textus hebraei, 3rd ed., Altdorf (1789)assigned Isaiah 40 onward to a later author writing in the exile
B. Duhm, Das Buch Jesaia, Handkommentar zum Alten Testament, Vandenhoeck & Ruprecht, Göttingen (1892)cut Isaiah in three: chapters 1–39, 40–55 and 56–66
M. Popović, M. A. Dhali and L. Schomaker, Artificial intelligence based writer identification generates new evidence for the unknown scribes of the Dead Sea Scrolls exemplified by the Great Isaiah Scroll (1QIsaa), PLOS ONE 16 (2021) e0249769the handwriting analysis indicating that two scribes probably copied the Great Isaiah Scroll, one half each
J. Wellhausen, Prolegomena to the History of Israel, translated by J. S. Black and A. Menzies, Adam & Charles Black, Edinburgh (1885)the classical form (1878) of the four-source documentary hypothesis, in English translation
J. E. Carpenter and G. Harford-Battersby (eds.), The Hexateuch According to the Revised Version, 2 volumes, Longmans, Green, and Co., London (1900)the verse-by-verse source assignments scored here as fixed hypotheses

← back to the web summaries