From Distant to Public Reading

1 Introduction: Size

“Some people have read more”, wrote Franco Moretti in one of the most quoted paragraphs in what later became a kind of manifesto of the Digital Humanities,

but the point is that there are thirty thousand nineteenth-century British novels out there, forty, fifty, sixty thousand – no one really knows, no one has read them, no one ever will. And then there are French novels, Chinese, Argentinian, American... Reading ‘more’ is always a good thing, but not the solution. (2000a, 55)

Clearly, at this early stage, Moretti had not thought of computers; he merely wanted to point out the limitations of close reading. And he chose to do so in a way that echoes Erich Auerbach’s remarks on the exact same issue, in the last article he published before his death nearly fifty years earlier ([1952] 1969). But in an age of growing data, it is no wonder that this paragraph caught the eyes of scholars who suggested various alternatives to good old human reading. From “Distant Reading” (Moretti 2000a) to “Macroanalysis” (Jockers 2013), from “Surface Reading” (Best, Marcus 2009) to “Algorithmic Criticism” (Ramsay 2011), from “Cultromics” (Michel et al. 2011) to “Cultural Analytics” (Manovich 2020), and beyond, the size – the size of what is known as “the great unread” (Cohen 1999, 23) – is presented as a problem. And while size, in and of itself, is not the only problem that the Digital Humanities are having to deal with, it is probably one of the most prominent ones. In effect, focusing on size widens the gap between Digital Humanities and the more traditional humanities. Because, for all the desire for reading that does move between closeness and distance – a position that many uphold (Hammond 2017) – in the end, as graphically described by Jean-Baptiste Michel et al.,

[i]f you tried to read only English-language entries from the year 2000 alone, at the reasonable pace of 200 words/min, without interruptions for food or sleep, it would take 80 years. The sequence of letters is 1000 times longer than the human genome: if you wrote it out in a straight line, it would reach to the Moon and back 10 times over. (2011, 176)

The scale chosen here to describe the size of the corpus is not accidental: it is difficult to think of a more challenging task, for a machine-less person, than that of occupying outer space.¹

But what if we were able to read the ‘great unread’? In our field, of modern Hebrew literature, Moretti’s description above, as well as the metaphorical journey to outer space, are simply not relevant. Although no one knows exactly how many Hebrew novels have been written – and we will return to this point later – we can calculate roughly how many have been published since the first one, Ahavat Zion (The Love of Zion), was published by Avraham Mapu in 1853.² And the number is not that big: the community of readers of modern Hebrew literature was quite small in the first decades of the genre’s existence. A new Hebrew language, allowing for essential daily communication that did not rely solely on the language of the sacred sources, had only just begun to emerge. Thus, from the mid-nineteenth to the mid-twentieth century, a few dozen novels, at most, were published, while the prestige of Hebrew literature was bestowed on two other of its forms: poetry and short stories. From the middle of the twentieth century onwards, the trend began to change. And although poetry and short stories continued to be written, the novel gained a significant place (both quantitatively and qualitatively), a process that reflected diverse cultural, national and political developments. From the establishment of the State of Israel (1948) until the time that these lines are being written, a few thousand novels have been published. So, all in all, the total number of Hebrew novels apparently does not exceed 8,500. While not an insignificant number, it is not an unwieldy one.

At this point in time, fortunately, it does not require getting to the moon and back. Many scholars, especially the oldest ones, have read a lot of the material. And from a computational perspective, as described by Marienberg-Milikowsky, it is a fairly convenient corpus: large enough for algorithmic reading, but not too large for human reflection (Marienberg-Milikowsky 2019). The question is, however, can we reverse the picture? Is it humanly feasible for us to re-read the great read, with mechanised reflection? How can we handle such a large-yet-manageable corpus? What are some of the methodological approaches we might use to reach beyond the canon? Alone, of course, we are not able to do that. But “some people have read more”... maybe they can help? Maybe ‘reading more’ is a solution?

In the subsequent sections of this contribution we turn to examine different methods and ideas that attempt to deal with a large corpus of texts. Following this, we will present Roman Mafte’ach (literally: roman à clef),³ a new digital approach we have developed aiming to deal with the Hebrew novel as a whole. We will describe the questions that such an approach can give rise to and look at both its possibilities and limitations. We will conclude this contribution with a section devoted to our initial findings, as well as with a few of our reflections and further suggestions. Setting our new methodology within the larger frame of Digital Humanities, we wish to offer it as another bridge between traditional literary studies and the digital age.

2 Option #1: The Humanities and the Digitized Text

Indeed, there are ways to deal with great corpora, and projects in the humanities are able to respond to the challenges posed by them. Such projects, varied in discipline, scope and technique, strive to include a large number of texts as their source material; they approach these texts as containing data, that is

a digital, selectively constructed, machine-actionable abstraction representing some aspects of a given object of humanistic inquiry. (Schöch 2013; emphasis in the original)

Thus, the machine, determining the dominance of quantitative methods, plays a crucial part in such projects. The text being digitised is (almost always) a condition, enabling this kind of research.

A prime example lies in a study that was carried out by Michel et al., presented in their oft-quoted article mentioned above, “Quantitative Analysis of Culture Using Millions of Digitized Books” (Michel et al. 2011). Aiming to observe cultural trends through quantitative investigation, the authors used 5,195,769 digitised books; according to them, this is about 4% of all books ever written (176). The term ‘culturomics’ was coined by them to describe this form of computational lexicography which, using Google Ngram Viewer, focuses on culture and on human behaviour. A similar methodological approach was used in many subsequent studies in many fields: history, media, linguistics and, more relevant to our study, literature. Thus, most projects in computational literary studies (CLS) – a relatively new branch of literary criticism and one of the main subfields of the Digital Humanities – have heavily depended, especially since the turn of the twenty-first century, on big corpora of digitised texts analysed by means of computational processing. This is clear, for example, in the publications of the Stanford Literary Lab founded by Moretti and Jockers in 2009,⁴ and in the work of many other scholars and centres as can be seen in leading DH journals such as the Journal of Cultural Analytics (CA) that, in five years, has become one of the flagships in the field.

The computational manipulations are many and different. Among others, topic modelling algorithms, one of the most popular tools in the field, were designed to identify groups of words that are often associated with each other, in pre-chosen digitised documents, providing “a compromise between full manual tagging and naïve word counting” (Mimno 2012, 9). Good examples of this are Jockers’ and Mimno’s analysis of 3,346 works of fiction from nineteenth century English literature, focusing on the authors’ sex and on gender-related themes (Jockers, Mimno 2013), and Andrew Goldstone and Ted Underwood’s observation of trends in literary theory and criticism (Goldstone, Underwood 2014) – a study which is based on more than 21,000 scholarly papers in literary history, written over the course of 120 years. Such solutions must be accompanied by additional techniques, of varying levels of technological and conceptual complexity: from the use of graph theory (Moretti 2011), through other non-naive word-counting applications such as TTR (type-token ratio), word clouds and more (Hammond 2016; Rockwell, Sinclair 2016), to manual annotation (Meister 2014) and alternative modes of data modelling (Flanders, Jannidis 2019) also adapted for literary analysis (Piper 2019) – the market of solutions is not small.

These options are available for any trained scholar who wishes to see cultural and stylistic trends in a large corpus of texts. The field is well established: this is almost a paved pathway. Though the prerequisite of the texts being digitised might be an obstacle, it is a technical (as well as legal) hindrance that is not insurmountable. And indeed, we have begun to enable such statistical measuring methodologies by archiving digitised versions of the Hebrew novel. Nevertheless, the platform we are proposing here is somewhat different. We will describe it in detail after surveying another, pre-computational, option.

3 The Hebrew Novel in the Eyes of Many

The Hebrew novel, being both big-enough and small-enough, can be seen as a case-study by means of which one can examine answers to these methodological and conceptual questions. Analysing it with algorithmic tools is a feasible task (even if we have to put a lot of effort into finding and developing NLP solutions for Hebrew, which, like other Semitic languages, is morphologically and syntactically different than European languages). Validating the results manually, even only partially, is a feasible task as well. But such a corpus may offer an even greater opportunity: limited as it may be, it is an entire literary field, rather than a fragment artificially created for research purposes. We have not determined how the corpus will be defined; we have started from the very reality of it as a fact, as a promising fact. And we want to read it before it becomes too big.

And yet, reading 8,500 novels is a challenge, particularly if we do not want to reflect on them through second-hand hegemonic and canonic historiographers and critics; particularly if we want to include all of them, and not only the canonical and studied ones; and also not just understand them as, say, 3,000,000 pages that need to be processed by digital and computational means, ignoring the fact that these pages are “fulfilled with meaning, only in the act of reading” (in the words of Witting 1977).

With these challenges in mind, we decided to turn to the public, and to experiment with a new type of reading, a new concept if you will: that of public distant reading. We realised that, if we turned to the public, to broad communities of readers (not necessarily scholars of Hebrew literature), we would be able to address all of these issues. Through public reading, we will be able to cover, potentially, the entire corpus; we will be able to see beyond the canon, beyond the classics, and beyond the best-sellers; we will be able to get rid of hierarchies, both within the literary sphere of different sub-genres and other divisions, and also between our readers; and, as a cherry on the top, we will be able to gather hermeneutic-based knowledge that can deepen, and sometimes even contradict, any non-human computational attempt to analyse the material at a glance. In other words, turning to the public is a way of ensuring a broader scope as well as overall diversity, without paying the price of a non-reflective sampling of the texts in the corpus.⁸ Moreover, by bringing in non-professionals as reliable readers, the project is able to bridge the gap between academia and the public, igniting public interest in what is often considered ivory tower academic scholarship, and locating the novel (and its study) where it belongs: in society.

The method chosen is quite simple, and is based on the following steps:

1. composing a comprehensive questionnaire that will be addressed to the general readership, expert and non-expert readers alike, and that will ask for as much information as possible (objective and subjective) about the Hebrew novel;

2. distributing the questionnaire among readers’ communities through electronic media, social networks and more;

3. collecting the questionnaires, sifting through the data and structuring it, then analysing it with statistical and visual tools;

4. validating fundamental objective elements that emerge from the questionnaires against more authoritative sources of information (see below);

5. at the same time, building a comprehensive digital corpus of the Hebrew novel, and analysing it with algorithmic tools based on the insights arising from an analysis of the questionnaires.

“Admittedly, contemporary distant reading is usually based on textual evidence, or on social evidence on dead people, rather than on questionnaires” writes Underwood in his “Genealogy of Distant Reading” (2017). And indeed, besides questionnaires-based studies in pedagogy (see, e.g., Miall, Kuiken 1995), we know of only one study that employed such a methodology: Janice Radway’s Reading the Romance (1984), a feminist, groundbreaking work of reader-response, centring on a community of readers using interviews and questionnaires. Yet the focus of Radway was, first and foremost, the readers, and the ways in which they understood romance novels, as opposed to the way that critics understood them (Underwood 2017). Although we can use our data also to learn about communities of readers, this is not the only outcome of our project, which aims at using reader responses as a means for the study of the novel. In a way, we are applying the conceptual promise of computational literary studies into the realm of human reader-responses, in order to have a better understanding of our chosen literary corpus: the Hebrew novel.

In Roman Mafte’ach, then, we use questionnaires to collect data on every novel originally published in Hebrew, from Ahavat Zion (The Love of Zion) published in Vilna in 1853 till today, both in Israel and elsewhere.⁹ Thus, the project is haphazard in its essence, as we ask that readers answer a questionnaire on any Hebrew novel they have recently read: both canonical and noncanonical novels; both popular and belletristic; novels published by well-known and commercial publishing houses, as well as ones that were published by independent, small ones; Hebrew novels throughout the generations, across the continents in which they were published. We assume that once we see the Hebrew novel as a distinct phenomenon, and once we gain enough data on Hebrew novels, we will be able to draw connections between different types of data concerning different aspects of the novel. These connections have the potential to power ideas and arguments, and can only be perceived through public distant reading, when approaching the Hebrew novel from above.

4 Initial Findings

From readers to novels: the findings below are preliminary by definition. They represent the first experimental stage in the gathering of material, allowing us to test the method: its effectiveness as well as its challenges. Given the length of the questionnaire and the complexity of the task our respondents agreed to take on themselves, we estimated and hoped that by now we would have data on no more than three hundred novels. We knew that some readers might be deterred by its length and complexity, but we have decided to maintain this relatively high bar as a threshold; after all, the questionnaires should be answered seriously and attentively. The answers, we assumed, would be enough to indicate initial trends and to reveal meaningful developments. And happily, within fourteen months, 525 questionnaires were collected, filled out by 229 readers, who covered in total 386 novels. What we have here is data on roughly 4.5% of the Hebrew novels ever published – assuming the total number of novels is 8,500, as indicated by our inventory. Good enough for a start. Consistently and persistently, we cover more and more sections of the Hebrew novel. We do not end here, of course: we are continuing to disseminate the questionnaires to as many readers as possible, including professional ones for more-specific reading tasks (as we explain below).

Preliminary as they were, the findings were astonishing: many scholars must have read more than 300 novels in their professional lives, guided by specific research questions or any other personal interest, but no one has seen them the way we, now, can see them. No one, as far as we know, could outline an empirical picture obtained from accumulating synchronously so much data that is both comprehensive and random – in the positive sense of the word. 

For example, none of them could say with as much clarity as we can now, that the variety of languages represented in the Hebrew novel encompasses within it intense cultural and historical tensions. It should be noted, for instance, that the most prominent language after Hebrew is English (and not Arabic, as one might have presumed); that Yiddish is ahead – quantitatively – of German; and that both of these are far ahead, quantitatively, of a distinctly Jewish language like Ladino [graph. 1]. At least in the case of languages in the Hebrew novel, even if a later picture is different, we can cautiously estimate that the difference will not be dramatic.

This figure, in itself, does not indicate any processes of development. The picture is static, unreflective of the history of, say, the emergence of English in the Hebrew novel, or the influence of other cultures (German, Russian, American, etc.), each in turn, on the Hebrew republic of letters. As the study progresses, we aim to show such developments. Moreover, since we asked the general public to fill out a questionnaire based on a novel that they have recently read, it is contemporary literature that is absolutely dominant in our database.  As the following illustration shows, among the 525 questionnaires that have been collected so far as part of the project, a powerful bias towards the literature of the last twenty years is evident [graph. 2a].

The timeline that emerges from the questionnaires should be examined against the timeline that emerges from the full authoritative list of 8,500 novels [graph. 2b]; then, what seemed biased is actually seen as more accurate, as the two graphics reflect the uneven distribution of novels over time, with a notable peak in the last decade.

On the other hand, this bias, which must, of course, be taken into account when analysing the results, has also accentuated what is a very real and practical challenge: the general public does not guarantee direct and equal access to all kinds of knowledge. Clearly, contemporary readers tend to read recently published novels more than old ones. A lot can be learned from this about the reading habits of our readers (are there any other parameters – not only the time of publication – that influence the scope of reading today?). However, since our primary interest is in the Hebrew novel as a whole, we are encouraging professional readers (students, colleagues, research assistants) to fill in the gaps in our developed timeline.

Focusing on the range of years that is most represented in our data, some intriguing phenomena begin to emerge, things to examine in more depth in later stages of the project: when do Hebrew writers (avant-garde, probably) begin to use non-textual elements as part of the plot itself, not as an ‘external’ decoration to it, and not as part of the graphic novel? The answer, according to preliminary findings is, not before the last decade of the twentieth century. How does the typical graphic structure of a page in a novel change over time? According to the same preliminary data, it seems like in the mid-eighties some diversity began to form, with the ‘classic’ page (divided into paragraphs) taken over by two inverted page-models, one compact and dense, and the other spacious and airy. According to our readers, these are also the years in which the dominance of the omniscient extradiegetic narrator shifted, making way for the diegetic narrator, who is involved in the plot to some degree.

These findings still require future comparisons with earlier periods of the Hebrew novel. But some findings became even more significant than expected, precisely due to the proliferation of contemporary literature in our database. Among other things, in our questionnaire we examine the extent to which contemporary forms of human communication (emails, SMS, social networks etc.) are reflected in the novel – and the result was quite low: strangely, even books of the twenty-first century find it very difficult to translate our ultra-technological daily life into a mimetic representation [graph. 3].

Moreover, as shown in the next figure, technology as a subject is almost non-existent in the data collected [graph. 4]. Not to mention the climate crisis, which, though a recurrent topic in the news at the beginning of our current millennium, is not reflected as such in the data. True, Hebrew writers are writing and publishing in the present; and yet according to our relatively rich data, it seems that they either do not see ‘the now’ or they do not give it an artistic mantle. The latter phenomenon was documented in contemporary novels in general, as brilliantly shown by Amitav Ghosh (2016).

Some of our data show, also, a clear picture of the correlation between the intergenerational relationship within the family, and the sex of the novel’s author. As shown in the graphics, women authors write more about daughters (and their relationships with fathers and mothers) while men write more about sons (and their relationships with fathers and mothers) [graphs 5-6]. It might not be surprising; we can assume that, in general, different people tend to write on what is like them. However, grounding this notion in facts, and being able to make correlations between this and other issues (for example: themes that women or men write on, the average page number for each, and more) can help to gain significant insights, and to make larger arguments about human culture in general.

Given, for instance, the fact that more novels are written by male than by female authors (as shown in our database), the overall result of the intergenerational relationships question is that in every parameter the male is the focus of literary attention during the history of Hebrew novel: consistently, more fathers-sons relationships are represented than fathers-daughters relationships; more mothers-sons relationships are represented than mothers-daughters relationships. And surprisingly, the exact same result is obtained in third-generation relationships as well: more grandparent-grandchild relationships are represented than grandparent-granddaughters, and more grandmothers-grandchildren relationships than grandmothers-granddaughters. 

Our task is a big one for traditional methods, a humble one for computational ones. Since it is big, the current picture is partial and modest. But since this picture is also a humble one, it has become fruitful in spite of the difficulties.

One difficulty lies in the very basic notion of trust. We often get asked: “do you trust your readers? How come?”. Going over each questionnaire, on some rare occasions we do find mistakes: one who filled out a questionnaire about a translated novel (whose original language is not Hebrew), or another who included a children’s novel (a sub-genre we have decided to leave out of the project). But such innocent mistakes are easily fixed by erasing the redundant data. So yes, in general we trust our readers, and the quality of data that we collect once and again vouches for them. In addition, the inventory mentioned above provides a source of validation, anchoring at least a part of the questions in our questionnaire.

The following is more challenging: we will happily receive multiple questionnaires on any one novel. Interestingly, and predictably, such questionnaires are almost never identical – at least in the more complicated and interpretation-dependent questions. Although this is quite confusing, we are not treating it as a threat to the methodology. True, we are extra-cautious when attempting to include this information in standardised graphs. But apart from this, we celebrate the difference, the multiplicity, going over the different readings and appreciating the smaller, as well as the larger differences, as they imply different readings, different points of view, different opinions. Indeed, as traditional, close and human reading is: you can never read the same novel the same way twice, you can never read it the way your friend does, the way your colleague does. Moreover, we see it as an opportunity to examine our questionnaire, as well as some of the concepts in narratology and in the study of Hebrew literature.

The final section of Moretti’s paradigmatic article includes a manifesto calling for intellectual attempts to reach the great unread by developing new, audacious methodologies. We quote this final paragraph here, as we see our project as an answer – even if only partial, as answers tend to be – to Moretti’s call.

Fantastic opportunity, this uncharted expanse of literature; with room for the most varied approaches, and for a truly collective effort, like literary history has never seen. Great chance, great challenge (what will knowledge indeed mean, if our archive becomes ten times larger, or a hundred), which calls for a maximum of methodological boldness: since no one knows what knowledge will mean in literary studies ten years from now, our best chance lies in the radical diversity of intellectual positions, and in their completely candid, outspoken competition. Anarchy. Not diplomacy, not compromises, not winks at every powerful academic lobby, not taboos. Anarchy. Or as Arnold Schoenberg once wonderfully put it: the middle road is the only one that does not lead to Rome (Moretti 2000, 227; emphasis in the original).

Radical diversity; anarchy; collective effort; diversity of intellectual positions; varied approaches. All of these – diversity, anarchy, varied approaches – are here, all are interesting. True, all of these do not make the study any easier, but once this messy data – again, in the positive sense of the word – is before our eyes, it is hard to ignore it.¹⁴ 

5 Conclusions

At first it might seem as if what has been proposed here is nothing but a technical solution to a conceptual problem: the problem of size. But we are hopeful that, on second glance, things will take on a different aspect, and it will become possible to realise the potential of our approach. In a way akin to algorithmic practices, dealing with literature indirectly, reading (or, unreading) it through the lenses of others, has the potential to miss direct contact with the text itself. But unlike in the case of those practices, whoever chooses the approach proposed here will not give up on human (unmediated) contact with the text. In other words, distant public reading actually returns the reader’s interpretive sensitivity to the picture. Here, we do not extract literary, thematic, or structural insights from word clouds, or from vector representations of words in an abstract space; rather, we listen to what many readers have to say – through our structured yet broad questionnaire – even if that information is ‘dirty’, contradictory and confusing. We do not take the human out of the humanities: those who read the novels are human beings (not machine-learning algorithms); those who make the combinations and ask the questions are human beings, too; those who interpret the data, who perform the distant reading that this project enables, are human beings who wish to see the field from a distance. These human beings harness the technological processing and analytics for this task. A balanced quantitative, computational approach helps us to ground our intuition in data – graphs, numbers, percentages – without leaving the readers and scholars outside of the interpretive picture. Therefore, we cannot – we must not – expect human insights to be clean. Along with the many novels, there will be many readers, and therefore also many questions and answers.

This is also the case with the scope of the multiple possibilities which our approach opens: it does not contradict, prevent or limit other attempts in the humanities. In fact, it may aid them: first, as a tool for detecting and sorting phenomena in literature, and second, on a broader level, it can serve as a reference to be compared with other approaches. Thus, it encourages ‘traditional’ scholars to continue to define a smaller corpus, based on the questionnaires; at the same time it encourages DH researchers to continue on their own way – without reading – but through a thorough comparison of their computational findings with what emerges from the questionnaires. Roman Mafte’ach suggests an on-going, additive, and changing discussion, one which is intellectual (and public), humanist (and proven), thought provoking (and grounded in facts).

We want to read as many novels as possible, but we have not reached world literature; we remain within the bounds of our language literature, which is seen also as national literature. However, within the boundaries of this national literature, we have chosen the broadest common denominator – that of language which has no actual boundaries.¹⁵ Since Hebrew literature has always been written in different centres around the world making contact with different cultures, one might say that the borders in question are, in a sense, loose: they are not limited to the actual territorial boundaries of the nation-state (which, in many cases, are artificial with respect to culture); and nor are they restricted to the national, ethnic or religious origin of the author. “Our philological home is the earth; it can no longer be the nation”, wrote Auerbach ([1952] 1969, 17) beautifully, acknowledging that the task of world literature, in a world of nation-states, physical borders and national literatures, is not an easy one.