Itay Marienberg-Milikowsky

In Alfred Döblin’s famous novel Berlin Alexanderplatz, a certain Franz Biberkopf rejoins the modern city after a prolonged incarceration, and is astonished by the relentless, alienating pace of change. Over time, Biberkopf gradually becomes entrapped in a net of forces stronger than himself, and his bewilderment is reflected in the splitting of his voice—or, maybe, the narrator’s voice—into two (if not more) contradictory points of view. Thus, the telegraph is described in one sentence as “astonishing, clever, tricky,” while in a subsequent sentence, we read: “It’s hard to get enthusiastic about all this; it functions, and that’s all” (Döblin, Berlin Alexanderplatz, 76).1

Revolutionary technologies were at the heart of the collective Talmud Blog series, which dealt with “the interface of Digital Humanities and the study of Rabbinic Literature,” an interface that is only now beginning to yield fruit.2 Despite some critical assessments to the contrary, these technologies are thankfully much less threatening than the ones that made Biberkopf so dizzy. Nevertheless, while I can identify with Michael Satlow’s notion in a 2017 post (“Digital Humanities and Rabbinic Literature”) that these technologies are indeed “astonishing, clever, tricky,” and that their use in Talmudic studies is highly promising, the other, split voice in my head tells a quite different story. It is not simply that I am rather more skeptical than Satlow (and many others) about the potential of these innovations to make real, lasting contributions to the relevancy and funding of the field, but more basically because what was true of the early telegraph is true also of our state-of-the-art digital environment: “It functions, and that’s all.” A careful assessment of this voice will be fruitful, as it impels us to act quickly to fill the gap between hypothetical ideas and true scholarly achievements. In what follows, I will try to explain this modest but necessary shift in attitude.

Let me begin from a seemingly distant point. Despite the assumption that distant reading has always been wedded to the digital age, it is worth remembering that Franco Moretti’s paradigmatic article “Conjectures on World Literature,” in which he coined this term, does not even mention computers. This is also true of his Graphs, Maps, Trees (2005), excluding an incidental note about “computational stylistic” at the beginning (4). Computers are mentioned in Moretti’s work for the first time only in “Style, Inc.” (2007). Of course, ever since he co-established the Stanford Literary Lab in 2010, he mentions them consistently. Even then, I think, computers are not a goal in and of themselves, but a tool for something else.3

That “something” may be called distant reading, but even that name is somewhat misleading. Intuitively, a distant perspective goes hand in hand with great size, and Moretti expresses his desire to cover “the great unread”; his latest studies certainly embody this desire in many senses. But in the predigital stage of his work, and even later, Moretti also occasionally expresses something of no less importance; namely, a desire for a systematic mode of reading, a narrow methodological investigation of textual appearances of well-defined, countable phenomena: “a unit of analysis” through which “all of literary history becomes a long chain of related experiments” (Moretti, “Conjectures on World Literature,” 61–62). As a result,

This kind of “reading,” however, no longer produces interpretations but merely tests them: it’s not the beginning of critical enterprise, but its appendix. And then, here you don’t really read the text anymore, but rather through the text, looking for your unit of analysis. The task is constrained from the start; it’s a reading without freedom. (61n19)

Against this background, it is not surprising to later find a confession regarding the phrase used in previous drafts of the article, “serial reading” (Moretti, Distant Reading, 44): serial, not distant, although the two are interrelated. We can understand why computers eventually entered the picture; but they were not there from the beginning. Emphasizing the consistency of a systematic analysis sounds quite banal to philological-historical ears; but within the original context of Moretti’s article—a discussion about literary theory and a specific (albeit large) corpus—it seems almost a provocative claim, for reasons that are beyond the scope of this chapter.

Inspiring as they are, we do not have to treat Moretti’s words as “The Law of Moses from Sinai.”4 Nevertheless, this opening meditation on his work provides new insights useful for studying rabbinic literature in the digital age. By placing quantitative perspectives at the center, we can again test where the study of rabbinic literature actually stands.

Michael Satlow (“Digital Humanities and Rabbinic Literature”) justifiably notes that the conditions for digital research in rabbinic texts are relatively good compared to other ancient literatures. We enjoy high-quality, impressively smart databases on a daily basis. They include, among others,

• the Bar-Ilan Responsa Project for digitization of almost every important rabbinic source ever printed;
• the Friedberg Project for digitization of Jewish manuscripts like those in the Cairo Genizah, including its subproject “Hachi Garsinan” for variations of the Babylonian Talmud;
• the integrated Cooperative Development Initiative website for digitization of early rabbinic sources, with a linked index of secondary literature;
• Ma’agarim, the historical dictionary of the Hebrew language;
• Sefaria, for open digitization of Jewish texts.

These esteemed collaborative efforts not only provide us with amazing access to a huge variety of texts and versions, but also supply wonderful tools for analyzing them.

However, there is a world of difference between digitization—in other words, turning the nondigital into the digital (for instance, turning a handwritten manuscript into a digital file)5—and what can be termed “digital hermeneutics” or, better still, “computational/quantitative hermeneutics,” meaning a reflexive use of computational tools for the purposes of interpreting a text, examining its poetics, or describing its place in the longue durée of literary history, all the while remaining aware of the unique aspects (not to say values) of the quantitative perspective. While digitization projects achieve public and academic appreciation—after all, who would not want to have digitally accessible texts?—it is hardly surprising that attempts to harness computational forces for reading endeavors, which are often claimed to be inherently subjective and speculative, meet with more suspicious opposition.

Considered in this light, what rabbinic studies is missing at the moment is not necessarily (or at least not only) new sophisticated databases, archives, editions, and the like, but something quite different: an extended collaborative effort to manage quantitative experiments in interpreting texts; a constructive critical dialogue; and, above all, a theoretical or conceptual framework that will imbue this effort with greater significance, without relinquishing an intensive dialogue with other currents in all related disciplines.

The good news is that we don’t have to wait for this to happen. There is no reason to speak only in the future tense, as is so often done within the digital humanities.6 Many options are available to us right now, in part thanks to accessible and powerful tools like Palladio for network analysis, Voyant for automatic text analysis, and my personal favorite, CATMA, for the integration of automatic text analysis and free, “undogmatic” manual annotation. This is even before we consider basic software like word processors and spreadsheets, which still have much to contribute. There is so much to do! In what follows, I outline the conditions that make this sort of research feasible, and suggest a few initial hypothetical examples.

First, the target audience of this approach is not a group of “digital humanists” in the narrow sense of the phrase: those gifted scholars with their feet firmly planted, at the same time, on both sides of the ocean that separates the “two cultures,” the humanist and the scientific (Snow, Two Cultures). The target audience members may know nothing about programming, nor about coding; they do not prepare texts for reuse with adaptation to preestablished digital standards: it is no part of their intellectual passion. In fact, even standard acronyms like TEI, XML, CTS, and NLP may seem to them strange and meaningless; and many of them have neither the money nor the time for ambitious academic start-ups that will provide them with a return on investment only after a long preparatory process, if at all. But they do feel at home in the so-called regular forms of Talmudic studies (or literary studies, in any language or cultural context—it doesn’t really matter); and they are ready to challenge old assumptions with new questions. Conversely, they are prepared to challenge new assumptions with old insights, and they should be willing to bridge what appear to be opposing paradigms.

Second, we have to temper our expectation for full automation, time-saving, or insights that arise directly from raw data “by pushing a button” (and let us put aside “objectivity,” a naïve pre-Kuhnian image of the process). Yes, digital tools do make a lot of Sisyphean work speedy and effective, and good visualization can lead to unexpected thoughts. But if we want numbers and graphs, attractive and beautiful as they are, to tell us something important, we have to help them, to arouse them; and for this, serious human effort is needed. Integrating the humanized into the computerized is neither a mistake nor a regression: it is exactly what makes this method so stimulating.

Third, quantitative investigation, constructed systematically, can shed light not only on a huge corpus but also on a single small text. When someone tells you that your corpus is not big enough for digital manipulation, do not believe him; it depends solely on what you are seeking to do. How many times have you reread, from a fresh point of view, a lyric song, a simple Mishnah, or even a complicated miniature Talmudic story, and suddenly found yourself facing something that—you could swear—wasn’t there before?7 The same is true for this new approach. Slicing a text with Voyant by carefully examining its word list and the interrelationships of those words affords us an opportunity to see the text through its “deformations” or, as one might say, “decompositions.” In this way we can, for example, evaluate the effects of a novel’s characters–or a gallery of sages in a midrash—based on countable measures, such as who is speaking, when he is speaking, how often, what he is relating (or at least what his vocabulary is)—and so on and so forth: a literal “deconstruction” of a text.8 The quest of Russian Formalism for a concept of the artistic “device” or “technique,” coined almost exactly a century ago by Victor Shklovsky (“Art, as Device”), now becomes, perhaps more than ever before, a realizable goal.

Distant reading, therefore, is most significant when integrated in a complex outlook that does not ignore the multidimensionality inherent in a concrete text or corpus. For this, we have all that is needed: “traditional” scholarship, tools, and texts.9

Let us take three of Michael Satlow’s aspirations. Here we reach issues that are at the heart of Talmudic research, but the proposed arguments could easily be applied to other fields as well. It would indeed be very impressive to get a comprehensive automated social network analysis of all the links connecting all the sages in the entire corpus of rabbinic literature. Internal sectioning options would take us even further; the anthological character of Talmudic literature—preserving, transmitting, and remodeling numerous traditions, attributed to several hundred sages of different times and different places—calls, no doubt, for network analysis.

However, although this task is potentially achievable with automated technology, we can already begin with semimanual pilot tests (using Palladio, for example), covering defined textual units from the Mishnah or the Talmud. Comparing networks of one such unit in the Mishnah and parallel texts like the Tosefta, or in the Babylonian Talmud and a parallel text like the Palestinian Talmud, would be a good start: these graphs can be created within a few days.10 A starting point such as this can indicate what else might be done with a wider lens, though it also has merit in and of itself. By heaping on more and more analytical layers—checking not only who is related to whom, but also in what ways are they connected, what contents are transmitted between them, what are the rhetorical patterns that shape their relationships, and so on—we arrive in the end at a rich map, half-computerized, half-manual, totally systematized, and, one hopes, interesting.

Such a map would be interesting not because it reveals interactions that have so far gone unnoticed, but for quite the opposite reason: because, as Willard McCarthy wrote, “models conceal when they reveal” (McCarty, Humanities Computing, 52). The map and the text will not overlap, and the gaps between them will demand an interpretive effort, integral to the workflow described above. An article published recently by Maayan Zhitomirsky-Geffet and Gila Prebor (“SageBook”) demonstrates this challenge. The authors propose an innovative method for a semiautomated creation of a social network of sages in rabbinic literature, based on structural analysis of lexical and syntactic patterns in the Mishnah. The network, evaluated and controlled by experts of Talmudic literature, is not very surprising: as the researchers point out, it is consistent with accepted conventions in traditional research. It does, nevertheless, open up a fascinating window into the rich variety of forms to represent agreement and disagreement in the text in question, and into the ability to quantify them. However, choosing a network model inevitably neglects another hero of the Mishnah, perhaps its greatest hero: the anonymous impersonal speaker (editor?), who is more present (quantitatively!) than any named character but inherently finds it difficult to be represented as a node in the social network.

The structure of the Talmudic sugya can be another representative example. Briefly, the sugya is the basic textual element of the Talmud.11 It is constructed by a process of dialectical discussion on one topic, more or less, moving from a well-defined start, through (sometimes complex) argumentative stages, to a well-defined (though often ambiguous) end. This process employs formalistic building blocks—some with narrative quality—reappearing repeatedly as the elemental structures of the sugya, shaping its variable theme. Talmudic scholarship shows special interest in this meaningful literary unit, seen justifiably as a key to the understanding of the Talmud as a whole. Once again, DH methods appear to be attractive.

Theoretically, there are many automated ways of rethinking the sugya. Syntactic-semantic analysis by “tree-bank” (following Satlow’s suggestion), or thematic analysis by topic modeling, could provide fascinating (and contradictory?) starting points—though they are, unfortunately, far beyond the average abilities of most Talmudists. Mapping Talmudic events, implied by the quasi-narrative building blocks of the sugya, would also not be easy: not only because “event” is a very complex category in narrative theory (see Hühn, “Event and Eventfulness”), but also because the fundamental “acts” in Talmudic literature are, most probably, speech-acts. But if we were to tie these missions to one of the main “traditional” questions in Talmudic studies during recent decades—the question of the weight, character, function, and effect of the anonymous material, which has an immense and influential part to play (Vidas, Tradition)—we would very soon find that we do not have any choice but to annotate manually, sugya after sugya, without ignoring the nondecisive interpretive questions that are inherent to the process. It is not a bad idea, rather the opposite: it is not hard to imagine a collaborative workshop (or seminar) dedicated to creating a “marked-up edition” of Talmudic chapters, pointing by tags—and not without reasonable hesitation—not only to different strata of the chapter but also, say, to different levels of abstract thinking and halachic conceptualization in the text, to different intellectual institutions represented by it, or to poetic changes in the stories woven into it, to mention only three topics that are arguably connected to the question of the anonymous layer of the Talmudic text, which includes statements that are not attributed to any of the sages of the Talmud and is sometimes referred to as the stam.

Here too, we do not have to remain in the hypothetical realm. I have tried to undertake such a task, without the help of a team, but with the help of CATMA and Microsoft Excel. I manually tagged large sections of the Babylonian Talmud to find out how stories of different kinds are repeated in the text sequence, copied from one context to another (Marienberg-Milikowsky, “Beyond the Matter”). The process took a very long time, but the surprising results justified the effort. First, many stories stubbornly refused to fall into the accepted categories of traditional scholarship, requiring a thorough update of the tag-set—that is, of the genre system. Second, the tag-set update enabled the development of a new theory—a theory of the narrativization of Talmudic discourse—with which I sought to explain how stories evolve from an element within the argumentation of the sugya, gathering to itself the formal means of organizing it.

There is still much work ahead of us. If one wants to understand, within the context of Talmud scholarship, what “undogmatic reading” is, one has only to remember the problem of materials presented as tanaitic (attributed to the Tanaitic period, up to the early third century) or amoraic (attributed to the Amoraic period, third to sixth centuries) at the superficial level (which can easily be identified by a computer), and the well-known fact that many of these are likely to be artificial editorial compilations, which can be identified only by a human interpreter (although automatization of this operation could be instructive). Challenges of this sort are naturally well known to anyone who deals with complex literature—literature that declares one thing, while saying something quite different.

Sugya after sugya, chapter after chapter, text after text—novel after novel?—exhausting but exciting. This is certainly one of the main lessons that “serial reading” teaches us: generalizations can and should be reached, but they must be based on (1) a proportional corpus, and (2) a method suitable for dealing with a wide variety of phenomena and data—in short, a convincing way to read “the great unread.” As I have shown above, and as new studies demonstrate (Münz-Manor, “Analog Piyyut in a Digital World”), it is not an eschatological mission. CATMA, for example, is designed to support such a project—in any text, in any language, under any interpretational framework.

These initial suggestions give rise to a great many questions, only a few of which can be noted here. What is the role of numbers in our literary or humanistic thinking? How does the move from representative examples to an analysis of corpus-wide trends modify our approach to the small-scale examples that are still undeniably important? How can we “translate” (or “operationalize,” to use Moretti’s word) abstract and flexible ideas into strictly defined “units of analysis,” or tag-sets for an annotation that would be fluid and systematic at the same time? Above all, what can be learned from the connections, occasionally surprising, between measurable elements on the text’s surface and our theoretical-conceptual insights regarding its depth?

In fact, leading scholars of other literatures have broadly dealt with these questions (for instance, Goldstone and Underwood, “Quiet Transformations”; Meister, “Toward a Computational Narratology”; Moretti, “Operationalizing”). This is not the place to discuss their solutions and to adapt them to rabbinic studies. To do so, we need many more case studies: in point of fact, as Satlow noted, we are still at the very beginning. Therefore, alongside introductions, hypothetical debates, and methodological workshops, we simply need many more experimental computational studies on different dimensions of Talmudic literature, which will prove, one hopes, that the effort pays off.12 Achieving this goal, however, does not have to wait for the World to Come. Considering that so many Talmudic texts have been digitized, and given that there are numerous tools that can already help us, it seems that the time has simply come to start working. “It functions, and that’s all.” This fact, in and of itself, does not make computational research less attractive; it merely bestows on it an attraction of a different kind.