In textual scholarship, conjecture is the proposal of a reading not found in any extant witness to the text. It is predicated on the idea that words are always signs of other words; the received text also harbors the once and future text. When A. E. Housman alters a line from Catullus's Marriage of Peleus and Thetis (c. 62-54 BCE) so that it reads "Emathiae tutamen, Opis carissime nato," instead of "Emathiae tutamen opis, carissime nato," he is practicing the art of conjecture.[6] Housman's small orthographic tweaks produce sizable semantic shifts: by relocating a comma and capitalizing “Opis” so that it is understood not as the Latin term for “power,” but as the genitive of “Ops,” the mother of Jupiter in Roman mythology, he substantially alters the meaning of the line: “Peleus, most dear to his son, is the protector of the power of Emathia,” becomes “Peleus, protector of Emathia, most dear to the son of Ops [Jupiter].” The emendation, as the fictional Housman explains in Tom Stoppard's The Invention of Love, restores sense to nonsense by ridding the language of anachronism: “How can Peleus be carrisme nato, most dear to his son, when his son has not yet been born?”  [Stoppard 1997, 37].

Expressed in grammatical terms, conjecture operates in the subjunctive and conditional moods.[7] Because its range of motion extends beyond the pale of the empirical, its vocabulary is replete with coulds, mights, mays, and ifs. Such a vocabulary reflects not caution — on the contrary, conjecture is a radical, audacious editorial style — but rather a refusal to settle for attested states of texts. In this it is opposed to the current wave of archival models of editing, which, in their rejection of speculative and inferential readings, opt for representation in the indicative mood.

Conjecture has at various moments in history held a place of honor in our repertory of editorial paradigms. Taken collectively, for example, Housman’s critical writings offer a magnificent apologia of divinatio, as conjecture is still sometimes called. The strength of conviction, the peremptory eloquence, the sheer depth and breadth of knowledge contained in those pages brook no protests about the bugbear of intentionalist editing or the death of the author. But at the same time, conjecture is a foundling, strangely lacking a well-defined history, theory, methodology. Discussion of it, which more often than not occurs incidentally, tends to take place not on an expository or historical plane, but on a homiletic or poetic one. Thus, for example, amid all his scientific discourse on compounded variational formulas, ancestral groups, successive derivation, and other arcane topics pertaining to recension, W. W. Greg falls almost quiet when he turns to the subject of conjecture: “the fine flower” of textual criticism he calls it, choosing metaphor over definition [Greg 1924, 217]. Treatments of conjecture characterized more by brevity than rigor are the norm: the process is hazily glossed metaphorically or else grounded in the successful critic’s abstract qualities of mind, which do not always lend themselves to analysis (e.g., intuition, judgment, confidence, insight, authority, charisma). For these reasons, conjecture has customarily been seen as either a practitioner’s art, the conspicuous absence of a meta-literature a symptom of the bottom-up approach it tends to favor, or a prophet’s.

The scholarly language can at times suggest the intervention of a supernatural agent. The classicist Robin Nisbet speaks of a “Muse of Textual Conjecture,” whom he playfully christens, appropriately enough, Eustochia.[8] As one response to the exquisitely difficult recovery mission that conjecture sets for its critics, Eustochia might be perceived as a sad contrivance to a misguided endeavor; a deus ex machina called in to artificially resolve all textual difficulties. But consider, for a moment, the bleak alternative: a world in which conjectural knowledge never was or could be an article of faith. Samuel Johnson’s plaintive verdict that conjecture “demands more than humanity possesses”  [Johnson 1765, ¶105] gives our imaginations something to work with; now envision that same despondency on a collective rather than individual scale. Instead of Erasmus proudly proclaiming the reconstruction of ancient texts the noblest task of all, we’d have only Stoppard’s Housman making a Faustian pact, born of despair, to suffer the fate of Sisyphus if only each time the stone rolled back he were given another fragment of Aeschylus [Stoppard 1997].

What does this general silence indicate about conjecture, its status in textual criticism, and its practice? E. J. Kenney’s position (which he officially co-opts from the nineteenth-century critic Boeckh, but which nevertheless seems to be prevalent among Classical editors) that the solution to the more “difficult cases” of conjecture “comes in a flash or not at all”  [Kenney 1974, 147] is tantalizing, but the idea of a lucid eureka moment — eerily similar to the romantic notion of inspired creativity (“At its inception, a [romantic] poem is an involuntary and unanticipated donnée”  [Abrams 1953, 24]) — shouldn’t curtail discussion; it should open the floodgates. And because the archival model holds conjecture in abeyance, an entire new generation of young textual critics who have cut their critical teeth on documentary precepts are all but forbidden at the outset to engage in it. For this reason alone conjectural criticism should inspire inquiry, but also precisely because it pretends to elude historical, cognitive, or technical understanding.

Would it be possible for conjecture to assume again a position of prominence in editorial theory? It would take a concerted inquiry into history, method, and theory. We would need to seriously engage the psychology of intuition, the AI and cognitive literature of inference, and the history of conjecture as divinatory art and scientific inquiry. We would need to keep apace with advancements in historical linguistics and evolutionary biology (two fields whose relevance to textual criticism is vastly under-appreciated) and the burgeoning psycholinguistic literature of reading. Indeed, we would reap many rewards from outreach to the linguistic community. We would need to dispel the myth that conjecture is a riddle wrapped in a mystery inside an enigma. A curriculum for conjecture would give us better insight into the method underlying the prediction — or retrodiction — of ancient readings of texts. Not least, we would need to work decisively to bring conjectural criticism into the 21st century. Because it has traditionally been described as a balm to help heal a maimed or corrupted text, conjecture is in desperate need of a facelift; the washed-up pathological metaphors long ago ceased to strike a chord in editorial theory.

So what would an alternate language of conjecture sound like?

We might, for starters, imagine conjecture as a knowledge toolkit designed to perform “what if” analyses across a range of texts. In this view, the text is a semiotic system whose discrete units of information can be artfully manipulated into alternate configurations that may represent past or future states. Of course the computing metaphors alone are not enough; they must be balanced by, among other things, an appreciation of the imponderable and distinctly human qualities that contribute to conjectural knowledge. But formalized and integrated into a curriculum, the various suggestions outlined here have the potential to give conjecture a new lease on life and incumbent editorial practices — much too conservative for a new generation of textual critics — a run for their money.

What do I mean by conjecture, then? Giovanni Manetti's pithy definition — “inference to the imperceptible” — is good insofar as it captures the cognitive leap (“inference”) from the known to the unknown.[9] But it fails to denote the temporal dimension of conjecture so essential to a field like textual criticism, which deals with copies of texts that are produced in succession and ordered in time. Moreover, inference is not a monolithic category but can be further subdivided into inductive inference, deductive inference, abductive, intuitive, probabilistic, logical, etc. Manetti's definition thus needs to be massaged into something more apropos to textual scholarship. For the purposes of this essay, I'll define it as follows: conjecture is allographic inference to past or future values of the sign. Allographic is a semiotic term given currency by the analytic philosopher Nelson Goodman in his seminal Languages of Art and refers to those media that resolve into discrete, abstract units of information — into what might be more loosely called digital units — that can be systematically copied, transmitted, added to, subtracted from, transposed, substituted, and otherwise manipulated. Conjecture as it is understood in this work is indifferent as to whether the mode of allographic manipulation is manual, cognitive, or mechanical. Allographic contrasts with autographic, or continuous, media. Examples of the former include musical scores, alphabetic script, pixels, and numerical notation; examples of the latter, paintings, drawings, and engravings. Painter and theorist Julian Bell summarizes Goodman like this: “Pictures differ from other sign systems, such as writing, by being continuities in which every mark is interdependent, rather than operating through a combination of independent markers like the alphabet. (In computational terminology, they are ‘analog’ rather than ‘digital’ representations.)”  [Bell 1999, 228]. Words are conventionally allographic, images autographic, though much important work in recent years has contested the validity of these distinctions and examined borderline cases, such as pattern poetry. Indeed, the last two decades of textual criticism have witnessed a wealth of scholarship promoting the text's bibliographic or iconic codes. Despite the virtues of such visual approaches to textuality (and there are many), conjectural criticism has been hamstrung by their success. The general argument in this essay is that conjecture flourishes in an allographic environment, not an autographic one. By “allographic inference,” then, I mean the considered manipulation or processing of digital signs with the goal of either recovering a prior configuration or predicting a future or potential one.

If we set aside for a moment the original terms of my definition (allographic inference) and focus instead on the terms used to gloss them (digital processing), then one of the salient points to emerge is that conjecture as formalized in these pages is semiotic in the first instance and computational in the second. Conjecture, that is to say, involves digital signal processing or, to use the more general-purpose term, computation.[10] Moreover, just as digitality, as a concept, extends beyond the boxy electronic machines that sit on our desktops, so too does computation. As David Alan Grier reminds us in When Computers Were Human, homo sapiens has been computing on clay tablets, papyrus, parchment, and paper for millennia [Grier 2005]. A computer is simply someone or something that systematically manipulates discrete, abstract symbols. As conventionally understood, those symbols are numbers, and the manipulations performed on them are addition, subtraction, multiplication, and division. But computers are text manipulators as much as number crunchers. The subfield of computer science that operates on textual rather than numeric data is known as stringology. Reflecting this broader emphasis, the symbols that are the focus of this essay are primarily alphabetic, but also phonemic and molecular. The only criterion for the symbols to function conjecturally, as already stated in our definition, is that they be digital, or allographic. The processes discussed here and throughout likewise depart from computational norms. They are not always or only mathematical operations, but also semiotic or grammar-like operations: transposition, deletion, insertion, substitution, repetition, and relocation. I will have more to say about them later, but for now let me summarize the four poles around which string computation has been organized in this project:

1. Input: allographic/digital (texts, words, letters, phonemes, molecular sequences)
2. Processor: human or machine
3. Processes or Operations: semiotic or grammar-like (transposition, insertion, deletion, relocation, substitution)
4. Output: textual errors (scribal copying); conjectural reconstructions or projections (textual criticism, historical linguistics, evolutionary biology); wordplay (poetry); auguries (divination); plain text or cipher text (cryptography)

This framework allows us to associate activities, behaviors, and practices that wouldn't otherwise be grouped together: wordplay, divination, textual transmission, and conjectural reconstruction, for example. It also allows us, as we shall see, to perceive a number of underlying similarities among a set of disciplines that for most of the twentieth century didn't take much interest in one another, namely textual criticism, historical linguistics, and evolutionary biology.

Historically, we have referred to the source of a message about the future as an “oracle” or “sibyl” or “seer”; within the contexts specified in this essay, we can alternately refer to that agent, whether human or mechanical, as a “computer.” From this standpoint, divination is not “knowledge in advance of fact” so much as knowledge that is a (computational) permutation of fact [Jonese 1948, 27]. Justin Rye, for example, author of the imaginary language “Futurese,” a projection of American English in the year 3000 AD, proposes that the word “build” will be pronounced /bIl/ in some American dialects within a century or more as a consequence of consonant cluster simplification [Rye 2003]. Underlying this speculation is a simple deletion operation that has been applied to a factual state of a word to compute a plausible future state. Transition rules of this kind are key ingredients in all of the examples of conjecture discussed in this project: the wordplay of Shakespeare's fools, soothsayers, and madmen; the puns of the Mesopotamian or Mayan priest; the “emendations” of textual scholars; the doublets of Lewis Carroll; the molecular reconstructions of geneticists; the projections of conlangers (creators of imaginary languages); the transformation series of historical linguists, textual scholars, and evolutionary biologists.

These devices and systems are unified by their semiotic and, in some cases, cognitive properties and behaviors: a pun resembles an editorial reconstruction resembles a speech error resembles a genetic mutation, and so forth. Each of these transitive relations is explored in the course of this essay. Consider, as a preliminary example, the analogy between textual and genetic variation. It is the allographic equivalence between the two — the fact that both are digitally encoded — that makes possible a strong theory of translation, allowing one system to be encoded into the other and, once encoded, to continue to change and evolve in its new state. Brazilian-born artist Eduardo Kac, who coined the term transgenic art to describe his experiments with genetic engineering, exploits this allographic equivalence in Genesis (1999), a commissioned work for the Ars Electronica exhibition. Kac's artist statement reads in part as follows:

The key element of the work is an “artist's gene,” a synthetic gene that was created by Kac by translating a sentence from the biblical book of Genesis into Morse Code, and converting the Morse Code into DNA base pairs according to a conversion principle specially developed by the artist for this work. The sentence reads: “Let man have dominion over the fish of the sea, and over the fowl of the air, and over every living thing that moves upon the earth.” It was chosen for what it implies about the dubious notion of divinely sanctioned humanity's supremacy over nature. The Genesis gene was incorporated into bacteria, which were shown in the gallery. Participants on the Web could turn on an ultraviolet light in the gallery, causing real, biological mutations in the bacteria. This changed the biblical sentence in the bacteria. The ability to change the sentence is a symbolic gesture: it means that we do not accept its meaning in the form we inherited it, and that new meanings emerge as we seek to change it.  [Kac 1999]

Kac’s elaborate game of code-switching is enabled by what Matthew Kirschenbaum calls “formal materiality,” the condition whereby a system is able to “propagate the illusion (or call it a working model) of immaterial behavior.”  [Kirschenbaum 2008, 11]. Because this model is ultimately factitious, succeeding only to the extent that it disguises the radically different material substrates of the systems involved, it cannot be sustained indefinitely. But at some level this is only to state the obvious: abstractions leak,[11] requiring ongoing regulation, maintenance, and modification to remain viable. This fundamental truth does not diminish the power or utility — what I would call the creative generativity — of models.

I first became sensitive to the convergence of conjecture and wordplay while studying Shakespeare as a graduate student. The ludic language of Shakespeare's fools, soothsayers, and madmen seemed to me to uncannily resemble the language of Shakespeare's editors as they juggled and transposed letters in the margins of the page, trying to discover proximate words that shadow those that have actually descended to us in the hopes of recovering an authorial text. The pleasure George Ian Duthie, a postwar editor of Shakespeare, shows in permutating variants — juxtaposing and repeating them, taking a punster's delight in the homophony of stockt, struckt, and struck; hare and hart; nough and nought[12] — finds its poetic complement in the metaplasmic imagination of Tom O'Bedlam in King Lear or the soothsayer Philarmonus in Cymbeline. Kenneth Gross points to what he calls Tom's “near echolalia” on the heath in the storm scene, the “strange, homeless babble” that “presses up from within Tom's lists, in their jamming up and disjunctions of sense, their isolation of bits of language[:] . . . toad/tod-pole, salads/swallows, wall-newt/water.”  [Gross 2001, 184]. There is something scribal and exegetical in Tom's babble, just as there is something poetic and ludic in Duthie's editing. And in the way both manipulate signs, there is also something fundamentally conjectural.[13] Conjecture, then, can be profitably understood by adopting a semiotic framework — expressed here in computational terms — within which we can legitimately or persuasively correlate the rules and patterns of conjectural transformation with those of mantic codes and wordplay.

Consider, too, the soothsayer Philarmonus, solicited by Posthumus near the close of Cymbeline to help decipher Jupiter’s prophecy, inscribed on a scroll that serves as a material token of Posthumus’s cryptic dream-vision: The soothsayer responds by translating keywords into Latin, isolating their homophones, and expounding their relations through puns and false etymologies: Language here is an anagrammatic machine. By sleight of sound, Philarmonus[14] morphs mollis aer (tender air) into mulier (woman) and cracks the code. “Every fool can play upon the word,” quips The Merchant of Venice’s Lorenzo 3.5.41. And so can every prophet. What the incident in Cymbeline highlights is that the “capaciousness of ear” with which Kenneth Gross attributes Hamlet applies in equal measure to Shakespeare’s soothsayers, clowns, fools, tricksters, madmen — and, I would add, editors [Gross 2001, 13]. It is significant, for example, that Posthumus should initially mistake Jupiter’s oracle for the ravings of a madman, for the madman and the prophet hear and speak in a common register. The fool, too, engages in the same digital remixing of sound and sense, and it is worth recalling in this context that Enid Welsford’s classic study of the fool’s social and literary history devotes an entire chapter to his dual role as poet and clairvoyant [Welsford 1966, 76–112]. It is not only titular fools (or madmen or prophets or clowns — the terms quickly proliferate and become functionally equivalent) but also de facto fools who, in King Lear for example, interface between wordplay and prophecy. The epithet of fool, like that play’s ocular and sartorial puns, is passed from one character to another, so that by the end of the play one is hard pressed to identify a single character who hasn’t tried on motley for size. By turns the fool, Lear, Kent, Cordelia, Gloucester, and Albany bear the brunt of the term, lending credence to Lear’s lament that his decrepit world is but a “great stage of fools”  (4.6.187).

Let me try to connect the figure of the fool, like that of the prophet, more directly to the motifs of error, wordplay, and conjecture. A self-styled “corrupter of words”  (3.1.34), Feste in Twelfth Night, for example, is fluent in the dialect of riddles and puns that constitutes the shared vernacular of Shakespeare’s fools. He subscribes to a philosophy of language that views the manipulation of sounds as a form of sympathetic magic — a process of world-making through word-making (and un-making). In the following exchange with Viola, the bawdy and jocular terms in which Feste expresses this philosophy should not distract us from the radical theory of the sign that underpins it: that there exists an instrumental link between signifier and signified, such that reconfiguring the one provides a means for fashioning, shaping, or foreshadowing the destiny of the other:

Viola:

Nay, that's certain; they that dally nicely with words may quickly make them wanton

Feste:

I would, therefore, my sister had had no name, sir.

Viola:

Why, man?

Feste:

Why, sir, her name's a word; and to dally with that word might make my sister wanton.

Considered within the context of these lines, “corrupter” denotes someone who has the power to pervert the world by deforming (“dallying with”) the language used to signify it. But Shakespeare mobilizes other meanings as well. In the domain of textual scholarship, the term corrupt has historically been used to describe a text riddled with copy errors. Because Feste’s cunning mispronunciations and nonsense words hover between intelligibility and unintelligibility, they closely parallel scribal error: “I did impeticos thy gratillity”  (2.3.27), Feste nonsensically proclaims to Sir Andrew, prompting one critic to wryly note that “words here are at liberty and have little meaning apart from that which editors, at the cost of great labor, finally manage to impose upon them”  [Grivelet 1956, 71]. The “great labor” to which Michel Grivelet alludes involves exploiting the digitally encoded phonological structure of language to conjecturally recover authentic words from corrupt ones: petticoat from impeticos, or gratuity (perhaps gentility?) from gratillity. Impressionistically, then, it can often feel as though the play’s textual “corruptions” — easily confused with its fictional vocal corruptions - were originating from within the story rather than from without; as if the text were inverting itself, such that the source of error were imaginary rather than real, with the fool becoming not only the object of a flawed transcription but also — impossibly — the agent of it. Because they mimic textual corruption, Feste’s broken, disfigured puns seem to compress temporality, exposing what the linguist D. Stein would call “diachronic vectors in synchrony.” [15]

Let me pause here to gather together a number of threads: my central tenet is that the linguistic procedures of Shakespeare's editors often seem to mimic those of Shakespeare's most inveterate “computers” of language. To emend a text — to insert, delete, and rearrange letters, phonemes, or sequences of words — is to ritualistically invoke a divinatory tradition of wordplay that dates back to at least the third millennium BC. For a benign alien power observing human textual rites from afar, the linguistic manipulations of a Duthie or a Housman would, I imagine, be for all intents and purposes indistinguishable from those of a Philarmonus or Feste or Tom O’Bedlam or Chaldean or, for that matter, a computer programmer using string-rewriting rules to transform one word into another. What all these examples have in common are digital units manipulated by computers (programmers, soothsayers, madmen, fools, punsters, poets, scribes, editors) using a small set of combinatorial procedures (insertion, deletion, transposition, substitution, relocation) for conjectural, predictive, or divinatory ends. Duthie and Tom O'Bedlam, computer programmers and Kabbalists, Philarmonus and historical linguists: they all compute bits of language.

As an editorial method, such alphabetic computation is dismissed by Housman as a wanton orthographic game. Finding a kindred spirit in a nineteenth-century German predecessor, he approvingly quotes the following:

Some people, if they see that anything in an ancient text wants correcting, immediately betake themselves to the art of palaeography . . . and try one dodge after another, as if it were a game, until they hit upon something which they think they can substitute for the corruption; as if forsooth truth were generally discovered by shots of that sort, or as if emendation could take its rise from anything but a careful consideration of the thought.  (Qtd. in Housman 1921.)

And yet the conjectures of, for example, Richard Bentley (1662-1742), one of the few textual critics for whom Housman professes admiration, are as much the product of letter play and combinatorial art as they are of careful deliberation and thought. (Given the structure and workings of the mental lexicon, discussed below, we might reasonably assert that letter play is the sine qua non of conjectural thought.) Lamenting the corrupt state of the blind Milton's Paradise Lost, which, according to Bentley, was not only dictated to a tone-deaf amanuensis, but also, adding insult to injury, later copy-edited by a derelict acquaintance, Bentley attaches a preface to his edition of the poem that includes a table of the “monstrous Faults” disfiguring the masterpiece, alongside Bentley's proposed emendations [Bentley 1732]. Reflecting the editor's conviction that “Words of a like or near Sound in Pronunciation” were substituted throughout for what Milton intended [Bentley 1732], the table reads like a dictionary of wordplay: “is Judicious” is emended to “Unlibidinous”; “Nectarous” to “Icarus”; “Subtle Art” to “Sooty Chark”; “Wound” to “Stound”; “Angelic” to “Adamic.” [16] The same “near echolalia” that characterizes Tom O'Bedlam's speech in the storm scene also characterizes Bentley's editorial method.

To adjudicate among variants, Bentley finds evidence and inspiration in diverse knowledge domains. Milton's “ secret top of Horeb” in the opening lines of Paradise Lost is conjecturally restored to “sacred top” through an appeal to literary tradition, geology, meteorology, and logic. According to Bentley, “secret top” finds little precedent in the works of antiquity, while “sacred top” has parallels in the Bible, Spenser, and various classical authors [Bentley 1732, 1n6]. Emending a text by bringing it into alignment with literary antecedents is a technique in which Bentley takes recourse again and again, as though a poem were best thought of as a commonplace book filled with its author's favorite quotations. The model of authorship that underwrites this approach is one that stresses sampling and collage over originality and solitary genius. The more allusive and intertextual the work, the more susceptible it will be to conjectural reconstruction.

Whatever the shortcomings of Bentley's appeal to precedent, it has the effect of helping systemize and guide his wordplay. His paronomastic methods are constrained by other factors as well: the double articulation of natural language, comprised of a first level of meaningful units (called morphemes) and a second level of meaningless units (called phonemes), imposes order and rules on the process. There are in English a total of twenty-six letters of the alphabet capable of representing some 40-45 separate phonemes, and phonotactic and semantic constraints on how those letters and sounds may be combined. The sequence “ptk” in English, for example, is unpronounceable, and the sequence “paf,” while pronounceable, is at the time of this writing meaningless, except perhaps as an acronym. Bentley's wordplay is thus bounded by the formal and historically contingent properties of the natural language with which he works, properties that theoretically prevent his substitutions and transpositions from degenerating into mere gibberish.

The importance of digital or allographic units to computation, as I have defined it here, is underscored by Gross's observation that what Tom O'Bedlam manipulates are “ bits of language” (my emphasis). Likewise, Elizabeth Sewell repeatedly emphasizes that nonsense poetry and wordplay require “the divisibility of its material into ones, units from which a universe can be built”  [Sewell 1952, 53]. She goes on to say that this universe “must never be more than the sum of its parts, and must never fuse into some all-embracing whole which cannot be broken down again into the original ones. It must try to create with words a universe that consists of bits ” (my emphasis) [Sewell 1952, 53–4]. It is precisely the fusion of parts into an indivisible whole that has made pictures — particularly dense, mimetic pictures — historically and technologically resistant to manual computation and, by extension, conjectural reconstruction.

The similarities between scribal and poetic computation (understood in the sense in which I have defined it) are brought home by even a casual look at the editorial apparatus of any critical edition of Shakespeare, where editors have traditionally attempted to ascertain whether a particular verbal crux is a poetic device in need of explication or a misprint in need of emendation. The Shakespearean text is one in which an error can have all the color and comedic effect of an intentional malapropism, and an intentional malapropism all the ambiguity and perplexity of an inadvertent error. Is Cleopatra's “knot intrinsicate” a deliberate amalgam of “intricate” and “intrinsic,” one that deploys “half a dozen meanings,” or, by contrast, an accidental blend introduced into the text by a distracted compositor? Is Hamlet's “sallied flesh” an alternative spelling of “solid,” one that also plays on “sullied,” or, more mundanely, a typesetting mistake? Does the Duke's “headstrong weedes” in Measure for Measure encapsulate some of the play's core themes, or is it an inadvertent and nonsensical deformation of “headstrong steeds ”?[17] We are in general too hasty, insists M. M. Mahood in Shakespeare's Wordplay, in dismissing these and other odd readings as erroneous variants rather than appreciating them for what they in many instances are: the artful “twists and turns” of a great poet's mind [Mahood 1957, 17].

That Mahood feels compelled to devote several pages to the problem of distinguishing between errors and wordplay in Shakespeare's text is worth lingering over. It suggests that there is something fundamentally poetic about the changes — or computations — a text undergoes as it is transmitted through time and space. This holds true regardless of how we classify those computations, whether as errors, creative interpolations, or conjectural emendations. It is as if a poem, at its most self-referential and rhetorical, were recapitulating its own evolution, or the evolution recapitulating the poem.

To clarify the point, consider the phenomenon of metathesis, the simple transposition of elements. Metathesis is a law of sound change (historical linguistics), a cipher device (cryptography), a poetic trope (classical rhetoric), a scribal error (textual criticism), and a speech error (psycholinguistics). There are entries for metathesis in both A Handlist of Rhetorical Terms (a guide to formal rhetoric) and A Companion to Classical Texts (a primer on textual criticism, which includes a typology of scribal error).[18] The overlaps between the two reference works are, in fact, considerable, with entries for epenthesis (the insertion of an element), homeoteleuton (repetition of words with similar endings) and other basic manipulations common to both. While it would be a stretch to claim that the lists contained in them are interchangeable, it wouldn't be much of one. Wordplay, one might conclude, is a poetics of error.

Cognitive evidence supports this view. Clinical experiments, slips of the tongue, slips of the pen, and the speech and writing disorders of aphasics (those whose ability to produce and process language has been severely impaired by a brain injury) all provide insight into the structure and organization of the mental lexicon. Current research models that lexicon as a web whose connections are both acoustic and semantic: like-sounding and like-meaning words are either stored or linked together in the brain [Aitchison 2003]. These networks of relation make certain kinds of slips or errors more probable than others. As natural language users, for example, we sometimes miss a target word when speaking or writing and instead replace it with one that occupies a nearby node, making substitutions like profession for procession or medication for meditation relatively common [Aitchison 2003, 145]. Transpositions like pasghetti for spaghetti or blends like frowl (frown plus scowl) are also legion and suggest that the unit of production and manipulation is the individual phoneme or grapheme: discrete, digital, and mobile [Aitchison 2003, 216]. For many aphasics, these errors completely overtake normal speech: clip might become plick; butter, tubber; or leasing, ceiling [Aitchison 2003, 22]. As Jean Aitchison observes, “the problems of aphasic patients are simply an exaggeration of the difficulties which normal speakers may experience”  [Aitchison 2003, 22]. But what is uncontrollable in the aphasic is deliberate in the poet. “There is a sense in which a great poet or punster is a human being able to induce and select from a Wernicke aphasia,” writes George Steiner in After Babel [Steiner 1998, 297]. To which I would add: or from everyday speech and writing errors.

Steiner’s insight repays further attention. Poets, like artists in general, often creatively stress-test the system or medium with which they work, probing its edges, overloading it, and pushing it beyond normal operational capacity. Discovering where language breaks down or deviates from regular use is the business of both poet and neuroscientist, providing a means of gaining insight into the mechanisms of language perception, processing, and production. But whereas the neuroscientist gathers data from aphasic patients in a clinical setting, the poet becomes, as it were, his own research subject, artificially manipulating the cognitive networks of meaning and sound that form his dataset. As Steiner notes, when intentionally created, ordered, and embedded in larger textual or linguistic structures, such distortions become poetry.

The mantic wordplay in Shakespeare's Cymbeline has forerunners in divination methods indigenous to cultures as diverse as those of the Quiché Maya of South America and the Sumerians of the ancient Near East. Mayan priests still practice a type of calendrical prophecy that involves uttering the words of a specific date and interpreting their oracle by punning on them [Tedlock 2000, 263]. Notably, the Quiché term for punning is sakb'al tzij, “word dice,” which in the context of Mayan daykeeping makes links among wordplay, divination, and games explicit [Tedlock 2000, 263]. Later, we shall examine the kinds of string metrics that one can apply to puns and textual variants to compute optimal alignment, edit distance, and edit operations. What these are and why one might want to compute them are the subjects of another section; for now, it is enough to note that these algorithms rest on the premise that words and texts change procedurally — by adhering to the same finite set of legal operations we've previously discussed.

We can also demonstrate the formalism of divinatory practice by looking at some of the earliest inscribed prophecies of the ancient Near East, which make extensive use of the same conditional blocks found in modern computer programs to control the flow and “output” of the mantic code:

If it rains (zunnu iznun) on the day (of the feast) of the god of the city — [then] the god will be (angry) (zeni) with the city. If the bile bladder is inverted (nahsat) — [then] it is worrying (nahdat). If the bile bladder is encompassed (kussa) by the fat — [then] it will be cold (kussu).  [Manetti 1993, 10]

The divinatory apparatus consists of an “if” clause (in grammatical terms the protasis), whose content is an omen; and the “then” clause (the apodosis), whose content is the oracle. We should be wary about assuming, as James Franklin does, that the divinatory passage from omen to oracle is an entirely arbitrary one (“all noise and no laws”  [Franklin 2001, 162]). As the examples make clear, the passage is allographically motivated, “formed by the possibility of a chain of associations between elements of the protasis and elements of the apodosis,”  [Manetti 1993, 7] specifically a phonemic chain in the examples given above. What we are looking at, once again, is a jeux de mots. One word metamorphoses into another that closely resembles it by crossing from the IF to the THEN side of the mantic formula.[19]

The divinatory mechanism may be construed as a program for generating textual variants: the baru, the divinatory priest or technician, inputs the protasis-omen into the system. A finite set of legal operations (substitution, insertion, deletion, relocation, transposition) is performed on its linguistic counters, which are then output in their new configuration to the apodosis-oracle.[20] Understood algorithmically, the Mesopotamian divinatory code is a proto-machine language, one that precedes by several millennia Pascal or Java or C++.[21]

The allographic operations used to generate a future text in Mesopotamian divination are formally consonant with those that produce variant readings in an open print or manuscript tradition. Mesopotamian divination simulates textual transmission, by which I mean that it seems to accelerate or exaggerate, as well as radically compress, a process that would normally occur over centuries or millennia — and one, moreover, that would in a great many of its particulars occur inadvertently rather than (as in Mesopotamian divination) deliberately. It is as if we were watching a text descend to posterity through time-lapse photography. These conceptual and behavioral associations are visually reinforced by the traditional emblems of the baru, which consist of a writing tablet and stylus.[22]

The scholarship on Mesopotamian omen sciences published within the last two decades suggests that divinatory systems in the ancient Near East exerted profound influence not only on the content of the literary genres of the ancient world, including the written record of the Israelites, but also on the interpretive practices of their exegetes.[23] The affinity between Sumerian divination and early textual scholarship of the Hebrew Bible is expertly established by Michael Fishbane in Biblical Interpretation in Ancient Israel, an extraordinary study of the scribal phenomenon of inner-biblical exegesis. Fishbane writes as follows:

Sometimes these [Mesopotamian prophecies] are merely playful jeux de mots; but, just as commonly, there is a concern to guard esoteric knowledge. Among the techniques used are permutations of syllabic arrangements with obscure and symbolic puns, secret and obscure readings of signs, and numerological ciphers. The continuity and similarity of these cuneiform cryptographic techniques with similar procedures in biblical sources once again emphasizes the variegated and well-established tradition of mantological exegesis in the ancient Near East — a tradition which found ancient Israel a productive and innovative tradent [i.e., transmitter of the received tradition].  [Fishbane 1989]

Fishbane’s work demonstrates through exhaustive research that mantic practice of the ancient Near East is a crucially important locus for understanding early textual analysis and emendation of the Hebrew Bible. Our working hypothesis must therefore be that the connections between these hermeneutic traditions are causal as well as formal. Historically, then, and contrary to popular belief, prediction has been as much a part of the knowledge work of the humanities as of the sciences — as much our disciplinary inheritance as theirs. It is time that we own that legacy rather than disavow it. We have for some time now outsourced conjecture to the Natural, Physical, and Computational Sciences. Paradoxically, then, it is only through interdisciplinarity that we can possibly hope to reclaim disciplinarity.

This section looks at the shared “arboreal habits,” to use Darwin's term [Darwin 1871, 811], of conjectural critics in three historical disciplines: textual criticism, evolutionary biology, and historical linguistics. More than a century after the publication of Darwin's On the Origin of Species, which helped popularize the Tree of Life, tree methodology remains at the center of some of the most ambitious and controversial conjectural programs of our time, including efforts to reconstruct macrofamilies of languages, such as Proto-Indo-European, and the Last Universal Common Ancestor (LUCA), a single-celled organism from which all life putatively sprang.[24] At the outposts of the biological and linguistic sciences, conjecture is prospective as well as retrospective: synthetic biologists and language inventors are known to experiment with modeling future states of genomes and languages, respectively.[25]

In an article entitled “Trees of History in Systematics and Philology,” Robert J. O’Hara, a biologist with impressive interdisciplinary credentials, draws attention to the presence of “trees of history,” glossed as “branching diagrams of genealogical descent and change,” across the disciplines of evolutionary biology, historical linguistics, and textual criticism. Figures 1, 2, and 3 after O’Hara reproduce three such trees [O'Hara and Robinson 1993, 7–17], all published independently of one another within a forty-year span in the nineteenth-century.

Figure 1 is Darwin’s well-known tree of life from On the Origin of Species (1859). Figure 2, the first manuscript stemma ever printed, diagrams the lines of ancestry and descent among a group of medieval Swedish legal texts (1827). And Figure 3 is a genealogy of the Slavic languages (1853). As an evolutionary biologist — a systematist to use the technical term — O’Hara’s interest in “trees of history” stems from his work in cladistics, a system of classification based on phylogenetic relationships in evolutionary groups of organisms. Over the last several decades, cladistics has become a highly sophisticated computer-assisted methodology, one that has outpaced stemmatics and its linguistic counterpart technically if not conceptually. It is unsurprising, then, that in addition to satisfying a general theoretical curiosity about the coincidence of trees of history in three seemingly disparate fields, O’Hara’s research is also an outreach project. He has, for example, collaborated with Peter Robinson to produce cladistic analyses of Chaucerian manuscripts [O'Hara and Robinson 1993, 53–74]. Such outreach has awakened both textual critics and linguists to the application of bioinformatics to their respective fields. But while O'Hara triangulates all three historical sciences, the link between textual scholarship and linguistics receives comparatively less attention than the links each of them shares with evolutionary biology. It is this third leg of the triangle that I want first to consider.

The evidence suggesting a direct historical relationship between stemmatics in textual criticism and the comparative method in linguistics is relatively hard to come by in the secondary literature, and what little of it there is, is circumstantial. The English-language textual scholarship has almost nothing to say on the subject,[26] although I suspect the situation with respect to German-language scholarship may be different, and the linguistic and biological literature for the most part falls in step. Standard accounts of the comparative method, as the genealogical approach is known in historical linguistics, are more apt to note the influence of Darwinian principles than textual ones. But a small cadre of linguists and biologists, following the lead of the late Henry Hoenigswald, have set about revising what we think we know about tree visualizations in the nineteenth century.[27] Hoenigswald, a historical linguist as well as a historian of linguistics, has managed to unearth interesting connections by operating at a granular level of analysis.[28] He succeeds where others fail because he has something of a detective mentality about him: his work demonstrates that to successfully trace the itinerary of ideas, one sometimes has to be willing to track the movements of the individuals who propagated them. Who knew who when, where, and under what circumstances? His lesson is that it is useful to think in terms of coteries and their membership if the objective is to map the spread of ideas. While other scholars write very generally about the diffusion of the arboreal trope in the nineteenth century, Hoenigswald — and perhaps we see the imprint of the genealogical method here — attempts to discover direct lines of influence.

Any account of the origins of the comparative method in historical linguistics must take into account the achievements of A. Schleicher (1821-1868), the founding father of the Stambaumtheorie, or genealogical tree model, by which the relations of the variously known IndoEuropean languages were set forth in a pedigree, and an asterisked hypothetical ancestral form postulated. Comparative method is a misnomer to the extent that it suggests comparison is the focal activity of the method, when in fact it is more like a means to an end. Comparisons of variants between two or more related languages are undertaken with the purpose of reconstructing their genealogy and proto-language. While the method is partially indebted to the Linnean taxonomic classification system, Hoenigswald is quick to point out a second, perhaps more significant influence: as a student, Schleicher studied classics under Friedrich Ritschl (1806-1876), who, along with other nineteenth-century figures such as Karl Lachmann and J. N. Madvig, helped formalize the science of stemmatology [Hoenigswald 1963, 5]. The affinities between textual criticism and historical linguistics are such that very minor adjustments to the definition of comparative method would produce a textbook definition of stemmatics:

Comparative method: Comparisons of variants between two or more related languages are undertaken with the purpose of reconstructing their genealogy and proto-language.

Stemmatics: Comparisons of variants between two or more related texts are undertaken with the purpose of reconstructing their genealogy and archetype.[29]

In order to better understand how an unattested text, language, or genome is inferred from attested forms, I want to briefly look at two competing classes of stemmatic algorithms: maximum parsimony and clustering. Broadly speaking, clustering or distance methods compute the overall similarities between manuscript readings, without regard to whether the similarities are coincidental or inherited,[30] while maximum parsimony computes the “shortest” tree containing the least number of change events still capable of accommodating all the variants. The difference is often expressed in evolutionary biology as that between a phenetic (clustering) versus a cladistic (parsimonious) approach. Here is how Arthur Lee distinguishes the two:

Cladistic analysis is sharply differentiated from cluster analysis by that which it measures. Cluster analysis groups the objects being analysed or classified by how closely they resemble each other in the sum of their variations, using statistical “distance measures.” Cladistic analysis, on the other hand, analyses the objects in terms of the evolutionary descent of their individual variants, choosing the evolutionary tree which requires the smallest number of changes in the states of all the variants.[31]

In cladistic analyses, the evidentiary gold standard excludes most of the textual data from consideration: only shared derived readings as opposed to shared ancestral readings are believed to have probative value. An ancestral reading, or retention, is any character string inherited without change from an ancestor. A derived reading, by contrast, is a modification of a reading inherited from the ancestor, often motivated rather than arbitrary from a paleographic, metrical, phonological, manual, technological, aesthetic, grammatical, cognitive, or some other perspective (for example, the Canterbury Tales scribe who, when confronted with the metrically deficient line “But tel me, why hidstow with sorwe” in the Wife of Bath's Prologue, alters “why” to “wherfor” to fill out the meter). When such readings are contained in two or more manuscripts, they are regarded as potentially diagnostic. Having made it through this initial round of scrutiny, they are then subject to a second round of inspections, which will result in exclusions of variants that have crept into the text via routes not deemed genealogically informative. The transposition of the h and the e of the when keyed into a computer is a mundane example. So pedestrian is this error that your word-processing program most likely automatically corrects it for you when you make it. In the terminology of Don Cameron, it is an adventitous rather than indicative error.[32] The stemmatic challenge is to screen out the adventitious variants, and successfully identify and use the indicative variants.

Phenetic or clustering analyses involve the use of a distance metric to determine degrees of similarity among manuscripts. The Levenshtein Distance algorithm, for example, named after the Russian scientist who created it, tabulates the number of primitive operations required to transform one variant into another. We can illustrate its application with one of the most notorious variants in the Shakespearean canon. In Hamlet's first soliloquy, we encounter the following lines: The first folio reads “solid flesh,” while the second quarto has “sallied flesh.” The Levenshtein or edit distance between them is determined as follows: let “solid” equal the source reading and “sallied” the target reading. The first task is to align the words so as to maximize the number of matches between letters:[33]

s o - l i - d | | | | s a l l i e d

Sequence alignment, as the method is called, is a form of collation. Identifying all the pairings helps us compute the minimum number of procedures needed to change one string into another. A mismatch between letters indicates that a substitution operation is called for, while a dash signals that a deletion or insertion is necessary. Three steps get us from source to target: a substitution operation transforms the “o” of “solid” into the “a” of “sallied”; an insertion operation supplies the “l” of “sallied”; and another insertion operation gives us the penultimate “e” in our target variant:

solid  — > salid
salid  — > sallid
sallid — > sallied

A distance of three edit operations (one substitution and two insertions) thus separates the two character strings. In clustering, the greater the edit distance between different readings, the more dissimilar they are assumed to be.

Although heterogeneous in their approach, both clustering and maximum parsimony are ars combinatoria. That is, both involve making assumptions about and/or modelling various orders, numbers, combinations, and kinds of — as well as distances between — primitive operations that collectively account for all the known variants in a text's transmission history. Both also take entropy for granted: they assume the depredations of time on texts to be relentless and non-reversible, with the result that the more temporally remote a descendant is from an ancestor, the more dissimilar to the ancestor it will be. In terms of our tree graph, this means that a copy placed proximally to the root is judged more similar to the archetype than a copy placed distally to the root.[34]

In biology, textual criticism, and historical linguistics, a transformation series (TS) is an unbroken evolutionary sequence of character states. The changes that a word or molecular sequence undergoes are cumulative: state A gives rise to state B, which in turn generates C, and so on. The assumption is that each intermediary node serves as a bridge between predecessor and successor nodes; state C is a modified version of B, while D is a derivation of C. The changes build logically on one another and follow in orderly succession. Consider, for example, a hypothetical (and admittedly idealized) textual series: head/heal/teal/tell/tall/tail.[35] Each word differs from the next by exactly one letter. The transitions are gradual rather than abrupt, striking a balance between continuity and change. “If we can discover such a transformation series,” writes H. Don Cameron, “we have strong evidence for the relationship of the manuscripts” in which the readings occur.[36]

Those readings might relate to one another in the vertical direction, in which case the task of the textual critic is to determine their order and polarity: Conversely, they might relate to one another as leaf nodes in the horizontal direction,[37] under which circumstances the editorial challenge is to infer vertical relationships: From this diagram, one can reconstruct plausible vertical variants, which recapitulate the horizontal: In this graph, each internal node generates two children, one of which preserves the reading of the parent, the other of which innovates on it. The innovations, or mutations, collectively constitute the TS and form a single evolutionary path through the tree.

Both distance and parsimony methods shed light on transformation series. The degree of difference between two leaf nodes, for example, reflects their degree of relationship: in our reconstruction, tail and tall, which differ by only one character, share a more recent common ancestor than tail and head, which differ by four.[38] The variants are also ordered so as to minimize the total number of mutations along the vertical axis, consistent with the principle of parsimony.

Now an admission: the source of the head/heal/teal/tell/tall/tail series is the nineteenth-century British author and mathematician Charles Lutwidge Dodgson, better known by the pseudonym Lewis Carroll, who uses it to illustrate not textual transmission or biological evolution, but rather a word game of his own invention, which he called “doublets” [Carroll 1992, 39]. Alternately known as “word ladders,” doublets is played by first designating a start word and end word. The objective is to progressively transform one into the other, creating legitimate intermediary words along the way. The player who can accomplish this in the fewest number of steps wins.

The congruity between doublets and biological evolution is the subject of an ingenious essay by scientist David Searls, entitled “From Jabberwocky to Genome: Lewis Carroll and Computational Biology.” [39] Searls begins his section on doublets by quoting the opening lines of Carroll's nonsense poem “Jabberwocky,” famous for its portmanteaux, neologisms, and word puzzles:

Twas brillig, and the slithy toves

Did gyre and gimble in the wabe;

All mimsy were the borogoves,

And the mome raths outgrabe.

“Jabberwocky,” was originally published in Through the Looking Glass (1871), Carroll's sequel to Alice's Adventures in Wonderland (1865). In chapter 6, the character of Humpty Dumpty, who comments at some length on the poem's vocabulary, interrupts Alice's recitation to observe of the first verse that “there are plenty of hard words there. ‘BRILLIG’ means four o'clock in the afternoon — the time when you begin BROILING things for dinner.” [40] “Upon hearing this explanation,” writes Searls, computational biologists “will of course feel an irresistible urge to do something like the following:”

b r o i l - i n g | | | | | | b r - i l l i - g

[Searls 2001, 340].

That “something” is sequence alignment, a preliminary step to computing the edit distance between the two strings and creating an edit transcript specifying the semiotic operations required to mutate one into the other. “Jabberwocky,” manifests the same metamorphic impulse that would later guide Carroll's invention of other word games. Of the gradated form of doublets in particular, Searls remarks that “the notion of the most parsimonious interconversions among strings of letters is . . . at the basis of many string-matching and phylogenetic reconstruction algorithms used in computational biology”  [Searls 2001, 341] — and, of course, in textual criticism and historical linguistics.

Interestingly, Carroll himself alludes to the resemblance between doublets and Darwinian evolution in an example that anticipates by more than half a century the linguistic and scriptural foundations of modern molecular genetics. He satirically bridges the evolutionary gap between apes and humans with a six-step transformation series:

APE — > ARE — > ERE — > ERR — > EAR — > MAR — > MAN

Extending Carroll's experiment to sets of words, Searls reconstructs hypothetical ancestors for eight leaf nodes:

Here the words at the bottom level are used to infer ancestors, but they might also be used to project descendants. They support conjectures that converge backward in time or diverge forward in time.

While Carroll initially conceived of doublets as a game of substitutions, involving the exchange of one letter for another, he later also admitted transpositions, as in the third step of the following series:

IRON — > ICON — > COIN — > CORN — > CORD — > LORD — > LOAD — > LEAD

And although deletions and insertions were never formally a part of doublet puzzles, they were central to another game exploring word generation, which Carroll dubbed “Mischmasch.” [41] Indeed, taken collectively, Carroll's games incorporate the full spectrum of grammatical operations on strings, almost as if they were intentionally designed to model patterns of descent with variation in cultural and organic systems.

What should we make of the fact that scientific theories of transmission and reconstruction can be so effectively illustrated with a parlor game? Recall that a correspondence between conjecture and wordplay was established earlier in this essay. There we discovered that a textual critic endeavoring to recover a prior text and a diviner attempting to decipher an oracle by signs were often united in their reliance on letter substitutions, puns, anagrams, and other permutational devices. Noting that the procedures underlying such word games were amenable to algorithmic expression, we labeled both the textual critic and the prophet who perform them “computers.” At the same time, we maintained that they were computers of a special sort, proficient in the semiotic processing of strings rather than the mathematical reckoning of numbers. Into their ranks we can also admit evolutionary biologists and historical linguists, for whom the grammatical operability of signs is no less germane.

In a post-industrial, Westernized society, an individual with the proper education who displays an aptitude for string manipulation might find gainful employment as an evolutionary biologist or historical linguist or cryptographer. In another milieu, that same individual might instead be inducted into the ways of the poet or prophet. The Mayan priest who practices calendrical divination through puns, the Mesopotamian baru who deciphers oracles through wordplay, and the Shakespearean soothsayer who interprets auguries by means of substitutive sounds and false etymologies find their 21st-century scientific complement in the figure of the computational biologist. Searls remarks that Carroll's doublet puzzles reveal “a turn of mind well suited to methodologies used in modern computational biology.”  [Searls 2001, 339] This statement has the effect of endowing what might otherwise be seen as an idle pastime with scientific rigor and purpose. But we could also flip the words around to observe that the computational methods of the biologist reveal a turn of mind well suited to wordplay. By thus inverting the language, we begin to see the poet in the scientist as well as the scientist in the poet. The contention of this poet-scientist is that all living things evolve through combinatorial play. The Book of Life, it would seem, is a volume of poetry, a dictionary of wordplay, and a compendium of oracles all bound together.

In this essay I have tried to advance the proposition that conjectural knowledge in the humanities is a manifestation of the inalienable human need to imagine what might have been or could be or almost was. This mandate of the subjunctive is beautifully conveyed by George Steiner in After Babel:

It is the constructive powers of language to conceptualize the world which have been crucial to man's survival in the face of ineluctable biological constraints, this is to say in the face of death. It is the miraculous — I do not retract the term — capacity of grammars to generate counter-factuals, “if”-propositions and, above all, future tenses, which have empowered our species to hope, to reach far beyond the extinction of the individual. We endure, we endure creatively due to our imperative ability to say “No” to reality, to build fictions of alterity, of dreamt or willed or awaited “otherness” for our consciousness to inhabit.  [Steiner 1998, xiii–xiv]

For Steiner, that “imperative ability” is more than a psychological or linguistic impulse; more, even, than a creative or ethical force; it is also a life instinct.

In the course of this work I have also tried to show that algorithms on strings — vital to conjecture — are as computationally sound as algorithms on numbers. We are so accustomed to thinking that punsters and textual critics somewhat imprecisely reshuffle symbols, whereas mathematicians and engineers methodically calculate them that it will perhaps take a class of scientists who process strings rather than numbers to convince us otherwise. We have looked at just such a group of professionals, computational biologists, whose work on the chemical “letters” of DNA resembles nothing so much as the parlor game of doublets or word ladders. And yet this form of wordplay is couched in the language of computation rather than poetry, vetted by biologists rather than poets or literary scholars, and published in scientific journals rather than poetry rags or literary monographs. Perhaps most significantly, the similarities between the methods of the computational biologist and the textual scholar — like those between the soothsayer and textual scholar — are not coincidental, but causal and historical.

One of my primary objectives has been to dispute the commonplace assumption that the inner workings of conjecture are ineffable and opaque. R. J. Tarrant, for example, writes that “conjectural criticism advances slowly and unsteadily,” proceeding at a “glacial pace” and is not amenable to “discussion of methods and trends”  [Tarrant, 121]. Remarks of this sort are scattered throughout the literature of textual scholarship: E. J. Kenney and Martin West profess similar sentiments, as does F. W. Hall, who remarks that, traditionally, conjecture “proceeds from no method and conveys no certainty”  [Hall 1913, 151]. My position, by contrast, has been that conjectural knowledge is produced in more determinate ways than is customarily acknowledged, which is to say through not only (or necessarily) the statistical processing of discrete symbols, but also their grammatical or semiotic processing. Stated more generally, there are reciprocal relations among conjecture, computation, and semiotics. This essay, then, has been written with the conviction that a cogent theory of conjecture is a desideratum of textual studies.

But what about conjectural methods in humanistic fields other than textual studies and linguistics that don't rely on systems of inheritance and variation? Take history, for instance, where the historical imaginary is our predominant tool for filling in the missing details of the past; or digital preservation, where the archivist must designate a primary user community for the cultural objects she preserves. By projecting the knowledge base of this user group into the future, the archivist makes decisions about what kind of descriptive metadata and contextual information to include in order to “render the object intelligible” to posterity.[42] If she misjudges the needs and attributes of this community — if her conjectural faculties falter — then the archive may be less meaningful, accessible, or useful than it would otherwise be. Are these two forms of conjecture so very different, either from each other or from those already described, as to generally make comparisons between them baseless? Or is it possible to create an integrated research narrative capable of accommodating a variety of conjectural projects, perhaps through recourse to a sophisticated typology of conjecture?

My own hunch is that there are deep structures common to many types of conjecture and new methods available to others that await discovery and experimentation. Our success in identifying and exploiting them, however, will depend in no small measure on how committed the digital humanities community is in coming years to building an information infrastructure that supports new models of scholarly communication and research. Patrick Juola, a computational linguist at Duquesne University, for example, is currently developing a prototype system for exploring corpus-based approaches to conjecture in the humanities. As Juola explains, his idea is to program the computer “to make essentially random emendations to texts and then rank them for plausibility using a set of agreed-upon criteria,” the underlying assumption being that the machine “is capable of devising and examining many more possible readings than even the most prolific human author, and even if 99.9% of the possible readings are flat gibberish, the one in a thousand may include interesting and provocative readings that human authors have missed”  (Patrick Juola, email to the author, 3 Oct. 2008.) Noting that this data-driven approach can be extended in a variety of ways, Juola emphasizes the versatility of the comparative framework he is proposing:

I could equally compare American and British literature, or 19th and 20th century literature, or first-person and third-person novels, or any other interesting category. Essentially, what the prototype I envision will do is generate (and test superficially) random conjectures like [Works in Category 1] use [Semantic category B] more than [Works in Category 2].  (Juola, email to the author, 3 Oct. 2008.)

The promise of Juola's corpus-based conjectural criticism, however, is predicated on the kind of mass digitization from which humanities researchers, as Christine Borgman emphasizes in Scholarship in the Digital Age, have thus far benefited very little, at least as compared to their scientific colleagues.[43] The data content layer, moreover, lacks the infrastructural support that libraries, institutions, and publishers have traditionally provided for the document content layer.[44] The problem is exacerbated in the humanities by the fact that the distinction between data and publications (or data and documents) is fuzzy to begin with [Borgman 2007, 219]. If we can create the kind of institutional incentives recommended by Borgman to develop, manage, and curate the data content layer, then perhaps a new practice of conjecture will follow.

I would like to express my gratitude to the following friends, colleagues, and mentors for the substantive feedback they provided on earlier versions of this work: Morris Eaves, Kenneth Gross, Matthew Kirschenbaum, Julia Flanders, Alan Liu, and Katie King. Their judicious comments and ideas enriched my thinking about the project and formed the basis for my revisions. I am also indebted to the anonymous readers for DHQ whose detailed suggestions helped strengthen the essay.