About

This paper originated in a presentation at Turkologentag 2018 in Bamberg, Germany, 19--21 September 2018. The computational analysis was first presented at the international workshop "Creating Spaces, Connecting Worlds: Dimensions of the Press in the Middle East and Eurasia" in Zurich, 31 October -- 2 November 2019. The final version is meant for submission to a special issue of "Geschichte und Gesellschaft".

The current stable draft of this paper is version v0.2 and contains unprocessed references to sources and secondary literature, cross-references, and CriticMarkup. To comment / review / annotate this version via hypothes.is click here.

Note that network plots currently make use of a computational transcription of Arabic into Latin script. The intended general audience is assumed to not be able to read Arabic. Since the latest version of macOS Mojave and Gephi 0.9.2 finally render Arabic script in network scripts, I shall supply a secondary set of Arabic plots.

Introduction

{>>research questions are still missing<<}

This paper presents the dual race to catch up with methodological and epistemological developments outside the realm of Middle East studies in the field of digital humanities and with the global networks of authors and readers enmeshed in late Ottoman periodicals.

The Arabic periodical press of the late Ottoman Eastern Mediterranean¹ has received scholarly attention since the early twentieth century. Yet, core questions concerning the intellectual history of the periodical press and the social history of periodical production are still unanswered. Early Arabic periodicals from the late nineteenth and early twentieth centuries, such as Butrus al-Bustānī's al-Jinān (Beirut, 1876–86), Yaʿqūb Ṣarrūf, Fāris Nimr, and Shāhīn Makāriyūs' al-Muqtaṭaf (Beirut and Cairo, 1876–1952), Muḥammad Kurd ʿAlī's al-Muqtabas (Cairo and Damascus, 1906–18/19) or Rashīd Riḍā's al-Manār (Cairo, 1898–1941) are at the core of formative discourses that still reverberate through the Arabic-speaking Middle East: the Arabic (cultural) renaissance (al-nahḍa), Arab nationalism, and the Islamic reform movement. Scholarly approaches to these periodicals have been, for a long time, both encyclopaedic and anecdotal as well as biased by an almost exclusive focus on Cairo and Beirut. Often compiled by authors who themselves were journalists, these works adhere to a specific political view of Arab nationalism.² After the publication of some union catalogues and indexes,³ focus shifted to the intellectual history of the naḥda and Arab nationalism from the 1970s onwards. The methodological focus of this work was---and to a large extent still is---on opinion pieces and editorials in a small sample of canonical journals while the historiography of the periodical press itself was relegated to the margins. Two noteworthy exceptions to this rule are Donald Cioeta's work on the history of Thamarāt al-Funūn and censorship in Ottoman Beirut⁴ and Ami Ayalon's extensive writings on the press in the Arab Middle East.⁵ Detailed studies on longer---if not entire---periods of individual publications are still lacking beyond Cioeta's unpublished thesis and Dagmar Glaß' book on al-Muqtaṭaf.⁶ In consequence, we still need to answer questions such as: What are the core nodes (authors, periodicals, other works) in the ideoscape of the late Ottoman Eastern Mediterranean? Is the geographic bias of Cairo and Beirut justified if we look at more than the "easily" accessible handful of monthly journals mentioned above? How would we need to re-write the intellectual history of the final decades of the Ottoman Empire if the myriad of papers and their contributors from places as far as Algiers, Basra or Aleppo were included? The answers to these questions depend on another set of questions: Who authored the majority of articles in any one periodical that did not carry a byline? Even if they were absent from the place of publication for prolonged periods of time, as we know they repeatedly were? Can we confirm the common---and untested---assumption that the proprietor or editor-in-chief mentioned in a journal's imprint or a newspaper's masthead just authored all the anonymous texts themselves? Is it true, as some suspect, that editors invented contributors to their journals from far-away places to impress their audiences?

This paper presents a first foray into computational approaches to these questions by adopting methods broadly summarised as distant reading, namely social network analysis and stylometry. Computational methods depend on digital corpora. I start by contrasting the hyperbolic promises of digitisation with the bleak reality of extremely limited access to unsystematically digitised facsimile corpora of Arabic periodicals from the late Ottoman Eastern Mediterranean. A socio-technical infrastructure built upon English and Latin script as the hegemonic technical paradigm, I argue, contributes to a neo-colonial divide between the abundance of digitised cultural artefacts of the Global North and the invisibility of almost anything beyond. This necessitates substantial corpus building efforts for many scholars working on texts in non-Western languages written in non-Latin scripts and severely limits the scope of our computational scrutiny.

I will introduce my corpus-building project OpenArabicPE as a framework to unite transcriptions of a small number of early twentieth-century periodicals from grey online platforms with digital facsimiles for the purpose of validating the former. Consequently, any corpus built {==with these affordances==}{>>better wording<<} and the dependence on the work of anonymous others will not be systematically tailored to the research questions one might have. After substantial modelling efforts, we are now able to generate and analyse bibliographic datasets as well as run first stylometric analyses on this corpus of some 165 periodical issues of c. 2.65 million words. This corpus is small if compared to the vast data sets available for the global north through Chronicling America, Trove Australia, the British Newspaper Archive etc. {>>add projects<<}, which gave rise to numerous distant reading projects.⁷ Although our approach can by no means be characterised as distant reading of big data, it is the first systematic attempt to empirically answer core questions for the nascent field of Arabic periodical studies,{--⁸--} which are in turn indispensable for a proper source critique if one wanted to employ these periodicals for the historiography of the late Ottoman Eastern Mediterranean.

This paper explores the late Ottoman ideosphere of the Eastern Mediterranean through the question of authorship and references to other periodical titles and the resulting intellectual, social and geographic networks.

{>>add paragraph summarising my results: The geographic realm must be broadened to at least include Iraq but also Sudan (as in the case of al-Zuhūr, which doesn't feature in this paper)<<}

The promised land of digitised Arabic periodicals

The better known and at the time widely popular Arabic journals of the late Ottoman Empire---unlike their smaller rivals and more ephemeral newspaper copies---do not face the ultimate danger of their last copy being destroyed in the current onslaught from iconoclasts, institutional neglect, and wars raging through Syria, Libya, Yemen, and Iraq. Yet, copies are scattered across libraries and institutions worldwide. This makes it almost impossible to trace discourses across journals and with the demolition and closure of libraries in the Middle East, they are increasingly accessible to the affluent Western researcher only.

A quick look at al-Muqtabas shall illustrate this point. A search in WorldCat for the nine volumes of al-Muqtabas will return six different bibliographic entries, the first of which has 13 variants (called "editions" in the context of WorldCat), pointing to 34 libraries. If one follows each entry to the holding library's catalogue, one will find that the large majority of collections is incomplete and that collections commonly combine original volumes, reprints, microfilms, microfiches and even photo copies. {>>comment on the geographic distribution of collections<<}

Digitisation promises an "easy" solution to the problems of preservation and access. Instant access to tens if not hundreds of thousands of digitised periodical issues from the late Ottoman Empire evokes the gold rush in the American west and many people imagine a promised land of instantaneous one-click answers to any question they might have. The public and many scholars expect to be able to put a computer to such diverse tasks as a keyword search: Show me all instances of the word waṭan (Arabic for "homeland", "nation") across the ideosphere of the early Arabic press between Morocco and Iraq from its beginnings until the World War I. Or a social network analysis: Show me the discursive field of authors and their texts and its changes over time. These are important and---for a variety of reasons, some of which will be discussed in this article---still open questions. Unfortunately, the eager student of digitised Arabic periodicals will immediately find tools, data and skills lacking.

{>>add footnote on the pitfalls of keyword search and loss of context through interfaces<<}

The first question we encounter in our attempt to track the network of authors and texts is to which extent can we submit digitised periodicals to computational analysis, or rather, what is the meaning of digitised and access? The answer is, of course, very different from periodical corpora in Western languages. It is sobering and will put off but the most enthusiastic readers: from large corporate and institution-backed platforms⁹ to grey online libraries¹⁰ digitised in the context of Arabic and Ottoman periodicals commonly means nothing more than the provision of digital facsimiles. Some platforms, such as Hathitrust and Cengage Gale, include search functions, but these are extremely limited. The former, Google-powered, platform is obviously dysfunctional for Arabic text if one has a look at the text layer. The latter makes grand claims of "Newly-developed optical character recognition software (OCR) for early Arabic printed script"¹¹ but does not make the text layer accessible to scrutiny. We will therefore never know the extent of false negatives (one can manually check for false positives by going through the search results). Beyond copyright, access to these digitised facsimiles is always restricted to viewing individual pages in web interfaces and limited by subscription fees and membership of specific academic institutions.¹² To computationally answer the above questions, however, one would need unrestricted access to truly digital editions---that is, machine-readable editions of the full text with embedded structural and semantic information and in a standardised exchange format.¹³

In the absence of digital editions, any meaningful computational analysis of the connections between authors, texts, and periodicals as a venue for publication and review requires access to reliable standardised bibliographic metadata as a bare minimum. Unfortunately, even this data is practically non-existent. This is due to ambiguity and incorrect data found in the original artefact; to lacking familiarity with the particularities of these artefacts among cataloguers, librarians and scholars; and to a software stack ill-suited for anything but Western concepts of dates and names and Western scripts.

{>>Add comment on faulty publication data as provided by the periodicals themselves; a) dates, b) names<<} Periodicals seem to provide no dating challenges since publication dates were conveniently recorded in a masthead. However, periodicals across the Arabic speaking late Ottoman Eastern Mediterranean made use of at least four calendars. Newspapers and journals provided dates in any combination of the Ottoman fiscal, or mālī calendar and the reformed Julian calendar as well as the better known Islamic hijri and Gregorian calendars. In addition to at least three different year counts, these calendars and their users also differed in their conception of the calendric day. Most retained the old notion of a day commencing at sundown, while others adopted alla franca time with 24 equinoctial hours and a date change at midnight.¹⁴ Unfortunately supplied dates from mastheads frequently neither matched each other nor the day of the week the paper was supposedly printed on. A particularly illuminating example in our corpus can be found in the masthead to issue 6 of the Damascene journal al-Ḥaqāʾiq that read "First of the holy month of Muḥarram 1229 [hijrī], 3 Kānūn Awwal 1327 [mālī], and 20 December (Kānūn Awwal) of the year 1911 [Gregorian]".¹⁵ This line contains at least four errors: First the hijrī date is set in the wrong century and should read 1329. Second, the days of the mālī and Gregorian dates got mixed up. The Gregorian calendar was thirteen days ahead of Julian calendars since the year 1900. The corrected mālī date should read 20 Kānūn Awwal instead of the third, but because the mālī years began in spring with the month of mārt (March), the year count should not have been incremented to 1327 with the beginning of the new hijrī year. Therefore the correct mālī date would have been "20 Kānūn Awwal 1326". The Gregorian date, finally, is even wrong after fixing the mix-up of days with the mālī date. Instead of Kānūn Awwal (December) it should be Kānūn Thānī (January) and since December has 31 days, the correct Gregorian date should ultimately read "2 (instead of the substituted 3 from the original, faulty, mālī date) January 1911". How should one record this bibliographic nightmare? And which date-calendar combination should be considered the authoritative one? What if recorded publication dates were fictional to simulate a regular publication cycle and should therefore be conceived of as issue numbers that have only limited relation to an actual date?¹⁶

{>>2) lack of knowledge among cataloguers<<} Any attempt to answer these questions relies on the affordances of available information systems, that is people and their skills, abstract concepts, and actual tools to record and retrieve these data points. But cataloguers, librarians and even specialists of the late Ottoman Eastern Mediterranean are frequently unfamiliar with calendric systems beyond the solar Gregorian and the Islamic lunar hijrī calendars. Mālī years are frequently misread as hijrī years, which introduces a margin of error of up to two years for the last decades before World War I.¹⁷ {>>3) short-comings of the software<<} Second, most software is unable to work with anything but Gregorian dates out of the box. Even if cataloguers were able to correctly establish the calendar used in a periodical's masthead, the computing infrastructure would not allow them to enter this date into the digital record.¹⁸ Finally, even when bibliographic data is internally kept in structured form, it is not commonly shared in a standard-compliant and machine-actionable format, such as MARC or MODS.¹⁹ A good example for this state of affairs is the British Library's otherwise excellent Endangered Archives Programme (EAP), which digitised periodical holdings of the al-Aqsa Mosque's library in Jerusalem. If we look at the fourth volume of the journal al-Muqtabas available through EAP, we find that bibliographic information is solely provided in unstructured plain text either through the web interface or the IIIF API.²⁰ Publication dates are provided as Gregorian months even though the cover clearly states that al-Muqtabas follows the "Arabic", i.e. Islamic hijrī, calendar and despite each issue reporting the publication date as hijrī month. Consequently, there is a dissonance between the facsimile and the bibliographic information. al-Muqtabas 4(1) recorded the month of Muḥarram 1327 aH in its masthead. Depending on the local observation of the moon in Damascus, the journal's place of publication, this month began around 27 January 1909. Should al-Muqtabas 4(1) therefore be considered the January or the February issue? The cataloguers at EAP clearly thought the latter or their cataloguing software did not allow for date ranges.

Even if we had perfectly reliable digital re-mediations of the bibliographic information found in the periodical issues themselves, the vast majority of articles would remain outside our analytical scopes, since publishers did not provide (meaningful) bylines---most articles in journals and newspapers from Baghdad, Beirut, Cairo or Damascus did not credit their authors.{>>The illegality of this behaviour is mentioned below<<} One approach would be to subject all articles to stylometric analysis for authorship attribution (more on this below) but this again presupposes truly digital editions.

Note that a the full text of a periodical is necessary but not sufficient for many analytical queries and distant reading. It is certainly insufficient for close reading. The full text of a periodical would be nothing but a string of words. But periodicals unite different texts of various genres from multiple authors. These texts are commonly grouped into issues and volumes and longer ones were frequently serialised and scattered across issues. Some of these texts will be reprints from other periodicals or first printed publications of much older manuscripts. Some of the texts are responses, etc. In order to make sense of the full text of a periodical for both humans and machines it has to be modelled.

Optical character recognition (OCR), the technology to convert an image into machine-readable text, has come a long way and even hand-written text recognition (HTR) is fairly successful at least for Latin script.²¹ Automatic recognition of Arabic script, however, is severely lacking behind for a variety of reasons beyond the scope of this paper.²² Despite promising developments with the application of machine-learning technologies to pattern recognition,²³ automatic conversion of images of early Arabic periodicals is hampered by three factors: first, all OCR technologies depend on training sets of "gold standard" transcriptions as ground truth; second, low-quality fonts, inks, and paper employed at the turn of the twentieth century will inevitably result in poor print quality; and third, text recognition depends on layout recognition and multi-column texts with various intersections of boilerplate, ads, etc. pose serious challenges. Consequently these texts can currently only be reliably digitised by human transcription.²⁴ Funds for transcribing the tens to hundreds of thousands of pages of an average mundane periodical are simply not available, despite of their cultural significance and unlike for valuable manuscripts and high-brow literature.

Grey online-libraries of Arabic literature, namely al-Maktaba al-Shamila (shamela.ws), Mishkat, Saydal-Fawa'id or al-Waraq, provide access to a vast body of (mostly classical) Arabic texts including transcriptions of unknown provenance, editorial principals, and quality for a small number periodicals. These grey "editions" lack information linking the digital representation to material originals, namely bibliographic metadata and page breaks, which makes them almost impossible to employ for scholarly research.

Since we do not have the resources to proof and correct these texts, we had the idea to build our own corpus by uniting transcriptions from grey online libraries with the digital facsimiles from other sources as a means to verify the quality of the digital text. Thus "Open Arabic Periodical Editions" (OpenArabicPE) was born in autumn 2015. {>>comment on the arbitrary composition of this corpus<<}

OpenArabicPE attempts at catching up

OpenArabicPE establishes a framework for open, collaborative, and fully-referencable scholarly digital editions of early Arabic periodicals. The guiding principles of OpenArabicPE can be summarised as accessibility, sustainability, credibility. Starting with the mostly Damascene periodicals al-Muqtabas and al-Ḥaqāʾiq, OpenArabicPE demonstrates that one can produce scholarly editions that offer solutions for most of the above-mentioned problems---including the absence of expensive infrastructure---through re-purposing well-established open software platforms and by combining the virtues of immensely popular, but non-academic (and, at least under US copyright laws, occasionally illegal) online libraries of volunteers on the one hand with academic scanning efforts as well as editorial expertise on the other.

Within OpenArabicPE we devise workflows and tools to transform digital texts from shamela.ws into an open, standardised file format (XML) following the Text Encoding Initiative (TEI)'s guidelines,²⁵ to generate bibliographic metadata, and to render a parallel display of text and facsimile in a web browser. We add light structural mark-up for articles, sections, authors, and bibliographic metadata, and link each page to facsimiles from various sources, namely EAP, HathiTrust, and archive.sakhrit.co.²⁶ The latter step, in the process of which we also make first corrections to the transcription, though trivial, is the most labour-intensive because page breaks were commonly ignored by shamela.ws's anonymous transcribers. This point needs to be emphasised: each of the c.8500 pages breaks in al-Muqtabas and al-Ḥaqāʾiq needed to be manually marked by volunteers in order to link facsimiles to the digital text and thus make the text verifiable for human readers.²⁷ So far Dimitar Dragnev, Talha Güzel, Dilan Hatun, Hans Magne Jaatun, Xaver Kretzschmar, Daniel Lloyd, Klara Mayer, Tobias Sick, Manzi Tanna-Händel and Layla Youssef have contributed their time to this task.

All tools and the editions are hosted on the code-sharing platform GitHub under MIT and Creative Commons CC BY-SA 4.0 licenses for reading, contribution, and re-use.²⁸ {>>mention collaboration with Leipzig and the integration of Muqtabas into CLARIN<<} As part of the editions we provide structured bibliographic metadata for every article in machine-readable format that can easily be integrated into larger bibliographic information systems.²⁹

With OpenArabicPE we argue that by linking facsimiles to the digital text, every reader can validate the quality of the transcription against the original. We thus remove the greatest limitation of crowd-sourced or grey transcriptions and the main source of disciplinary contempt among historians and scholars of the Middle East. Improvements of the transcription and mark-up can be crowd-sourced with clear attribution of authorship and version control using .git and GitHub's core functionality. Such an approach as proposed by Christian Wittern³⁰ has recently seen a number of concurrent practical implementations such as project GITenberg led by Seth Woodworth or Jonathan Reeve's Git-lit.

Network of authors and texts

With major work on modelling on three journals from Baghdad, Cairo and Damascus---Lughat al-ʿArab, al-Muqtabas and al-Ḥaqāʾiq---done, we can now begin to submit this corpus to initial analyses of the social and geographic networks of authors and their texts as well as the network of periodicals referenced and cited in these three journals.

OpenArabicPE's corpus

The corpus comprises the full text of each issue of Lughat al-ʿArab, al-Muqtabas and al-Ḥaqāʾiq and a transcription of article titles and bylines for one volume of al-Ḥasnāʾ until the end of World War I.

Muḥammad Kurd ʿAlī (1876–1953) was the best known and, after the Young Turk Revolution of 1908, the most influential journalist and intellectual in Damascus. Before establishing his journal al-Muqtabas (The Digest) in Cairo in 1906 and the first daily newspaper to be published in Damascus in 1908 (also confusingly called al-Muqtabas), he had held minor government offices and worked at various public and private presses and periodicals in Damascus and Cairo. He was well-acquainted with leading figures of the Islamic reform movement in Egypt and Greater Syria. He was a member in Ṭāhir al-Jazāʾirī's "senior circle" in the early 1890s in Damascus and later moved and worked in Rashīd Riḍā's and Muḥammad ʿAbduh's circles in Cairo. After the Young Turk Revolution, Kurd ʿAlī returned to his hometown and the publication of al-Muqtabas moved from Cairo to Damascus in the journal's third year. In Damascus, al-Muqtabas soon became "the boldest, most coherent, consistent and committed proponent of reform and modernity [...] prior to World War I".³¹ Due to conflicts with the authorities over the reprint of a poem, Kurd ʿAlī again fled Damascus for Cairo and Europe in 1912. Consequently, al-Muqtabas was published from Cairo for a couple of months before Kurd ʿAlī was allowed to return once again. During World War I and Cemal Pasha's infamous term as commander-in-chief of the 4th Army and governor general of Syria, Kurd ʿAlī was able to win his support. He thus escaped the fate of Shukrī al-ʿAsalī{>> his co-editor at al-Muqtabas<<}, ʿAbd al-Ghanī al-ʿUraysī and other journalists from Beirut and Damascus, who were publicly executed on charges of treason. Similarly to their editor, the journal and the newspaper al-Muqtabas survived and continued publication until the final days of the war---albeit in shorter and less frequent editions due to material shortages. After the end of the war and the disintegration of the Ottoman Empire, Kurd ʿAlī abandoned the monthly and left the editorship of the revived daily newspaper al-Muqtabas to his brother Aḥmad. He founded the Arab Scientific Academy whose president he became in 1919 and served twice as Minister of Education (1920–22, 1928–32) during the French Mandate over Syria.³² The edition of al-Muqtabas gathers all 96 issues published between 1906 and 1917/18 with a total of some 7.000 pages and almost 2 million words. {>>comment on print runs<<} Issues consist of longer independent articles and sections with shorter articles such as brief reports on new discoveries and book announcements. Issues and pages differed widely in length. On average, each issue contained 65 pages with almost 300 words or 20354 word per issue.

Much less is known about the second Damascene journal in our corpus and the people behind it. The ʿālim ʿAbd al-Qādir al-Iskandarānī (1875--1943) from the notable al-Kaylānī family, who were closely associated with the Qadriyya Sufi order, published three volumes with a total of 35 issues of al-Ḥaqāʾiq (The Facts) in Damascus between 1910 and 1913. al-Ḥaqāʾiq was a periodical of the conservative Muslim establishment, who called themselves mutayyinūn (the very pious). This becomes clear from the content of al-Haqāʾiq with its focus on Islamic topics, its constant critique of Salafism and "Westernised" journals, such as al-Muqtabas, and its makers' apparent lack of familiarity with the by then established layout conventions of Arabic journals.³³ Unlike Kurd ʿAlī, al-Iskandarānī did not claim a single byline in "his" journal or any other periodical in our corpus. Pages contain much less text and issues are much shorter than al-Muqtabas with a total of 300186 words across 1436 pages (as some 17 pages are missing from the transcription, the word count is too low). Save for a double issue, the vast majority of issues (28 of 36) comprised 40 pages with an average of slightly more than 200 words each. This results in an average 8516 words per issue compared to the 20636 words of al-Muqtabas.

{>>Basic description of Lughat al-ʿArab<<} The Carmelite Father Anastās Mārī al-Karmalī (born Buṭrus ʿAwwād, 1866--1947) from Mt. Lebanon established the monthly journal lughat al-ʿArab (The Language of the Arabs) in Baghdad in 1911. The journal published 34 issues until its final number in June 1914. It is not clear to which extent al-Karmalī was involved as editor. Only a (small) handful of articles in lughat al-ʿArab carried his byline. Kāẓim al-Dujaylī (1884--1970), a self-taught journalist and poet and student of the Baghdadi Salafist Shukrī al-Alūsī and al-Karmalī, joined Lughat al-ʿArab as editor(-in-chief?). Similar to al-Ḥaqā'iq, issues comprised mostly 40 pages with an average of almost 240 words per page and a total of 11274 words.

The quality and significance of the analysis of bibliographic data is directly dependent on the quality of the information provided by the periodicals themselves and of our mark-up in the TEI source files. All relevant personal and place names in bylines and other source information must be marked up and linked to local and external authority files to allow for the necessary disambiguation and grouping of entities that might be referenced by multiple names in our corpus, such as "Rashīd Riḍā" and "Owner of al-Manār". In addition, by linking references to external authority files and the semantic web, we can harvest additional information on authors and locations, namely the geolocation for toponyms, transcriptions into Latin script necessary for working with most visualisation tools, and life dates for persons in order to establish the age of authors upon publication.³⁴{>>Comment on the issue of many visualisation tools, such as R and Gephi for macOS, not working with Arabic script<<}

Evaluating the corpus: network of referenced periodicals

Knowing that we work with a corpus whose composition is the result of external and unknown decisions by the contributors to al-Maktaba al-Shamela as to which periodical to transcribe, we can evaluate the performance of this corpus in representing the larger ideosphere of the periodical press in the late Ottoman Eastern Mediterranean by looking at the network of referenced periodicals. Explicit references to periodicals indicated by "jarīda XYZ" or "majalla ABC" were automatically marked-up using XSLT and regular expressions and linked to local and external authority files for disambiguation and additional bibliographic information. We then counted the references to each mentioned periodical and plotted the result as a network graph. We plotted networks of references by issue to account for the varying length of articles in each journal.{>>sequencing into even-sized chunks might be a good idea for statistical validity<<} In the network plots below each node signifies a periodical (newspapers and journals are not distinguished at the moment). Edges are drawn between periodical titles when one references the other. The thickness of the edges and the size of the nodes indicate the number of issues that reference a periodical (weight). Colours signify the {==in-degree==}{>>i.e. the number of edges that connect to a node;<<} or the number of journals in our corpus that mention this periodical (3 = orange, 2 = green, 1 = purple).

The first observation, common to all social networks, is that only a very small number of nodes are of relative importance, as measured in in-degree (number of edges connecting to a node) and weight of the edges connecting nodes. Out of a total of 465 different periodical titles, 421 or c. 90% were referred to by only a single journal. 344 periodicals are only mentioned in a single issue and 335 in a single article. The second figure shows the 44 core nodes---periodicals that were referenced in more than one journal---in more detail. Only 9 (2,13%) were referenced by three journals in our corpus. {==They are: al-Manār, al-Muqtaṭaf, al-Hilāl and al-Ḍiyā from Cairo, al-Muqtabas itself, al-Mufīd, al-Waṭan and al-Ḥaqīqa from Beirut and al-Ḥuqūq from Mt. Lebanon ==}{>>comment: this fits the standard narrative of important journals, with the exception of al-Mufīd<<} The centrality of the {==three==} Cairene periodicals, al-Manār, al-Muqtaṭaf, al-Hilāl*, which were all published by Syrian immigrants, tentatively confirms the standard narratives of the Arabic press.³⁵ If we had the means to construct our own corpus without the severe limitations alluded to above, these would be the journals to digitise. Second, this network is highly centralised in terms of geographic distribution. The 44 core nodes were published in only a handful of locations: Beirut (9), Cairo (7), Baghdad, Damascus, Paris (3), Alexandria, London, Mt. Lebanon, Saida and Zahle (1).{>>plot this information on a map?<<}{>>how many at which locations?<<}

A third observation of the larger network is that al-Muqtabas accounts for the vast majority of references to other periodicals by some orders of magnitude even after we account for al-Muqtabas having almost thrice as much issues as either al-Ḥaqāʾiq or Lughat al-ʿArab. {==If we assume that we haven't missed a significant number of references==}{>>since mark-up was done automatically, this seems unlikely<<}, then al-Muqtabas was more outward-looking and more involved in larger discourses of the day. Fourth, a closer look at the core nodes in the network reveals that all periodicals were primarily self-referential---indicated by the thickest edges connecting a journal to itself (for the purpose of this visualisation and to prevent circular edges, source and target nodes were separated). Fifth, the core nodes include number of surprises: al-Jinān was published by Butrus al-Bustānī and later his son Salīm al-Bustānī in Beirut between 1876 and 1886. This means that either al-Jinān was still relevant for certain discourses long after it ceased publication or that the corpus, spanning the years 1906 until 1918, contains a number of historiographic texts mentioning important journals of the past. Articles in Ibrāhīm al-Yazījī's al-Ḍiyāʾ, published in Cairo between 1898 and 1906, were equally referenced after the end of this periodical. Finally, the group of periodicals referenced in two journals in our corpus comprises a number of foreign titles such as Le Temps, Revue des Revues and Revue du Monde Musulman from Paris and The Times from London (alTan, mjl0 almjlat, altyms in the plots).

{>>what are the lacunae in this network? which periodicals are not referenced?<<}

analysis of metadata: network of authors

Sketching a network of periodicals and the references between them is only one part in the endeavour to layout the ideosphere of the late Ottoman Eastern Mediterranean. Another is the network of authors who published in these periodicals and the geographic distribution of places mentioned in bylines. {>>Other layers of this ideosphere not covered in this short overview but worth exploring are the network of works reviewed and mentioned in our corpus. What was read and written about in Damascus and Baghdad?<<} Knowing the importance of certain authors for an individual periodical is the basis for mapping the {>>social<<} network of authors across the late Ottoman ideosphere.{>>these findings would need to be contextualised by traditional intellectual history<<} The aim would be to map such a network for the hundreds of journals and newspapers published between Alexandria and Aleppo, Jaffa and Basra. Nevertheless, this method already provides valuable insights using our small corpus of bibliographic metadata:

It is worth going back to the bibliographic metadata, its shortcomings and the resulting consequences for our analysis before looking at this aggregated information. We are particularly concerned with the number of articles that carried bylines or otherwise easily identifiable authorship information. All journals in our corpus, like any other periodical at the time I have seen, seriously violated the relevant Ottoman press codes that required identifiable author information being published alongside each article.³⁶ In this regard al-Ḥaqāʾiq is the least offending. About 42 per cent of all articles (163 of 389) carried authorship information. Second is al-Ḥasnāʾ with 36 per cent (63 of 173), followed by Lughat al-ʿArab with 16 per cent (152 of 939) and al-Muqtabas with not even 13 per cent (377 of 2964). The picture becomes more favourable for the latter if we look only at longer articles. About two fifths of authors outside news and review sections explicitly mentioned an author. However, in consequence and due to the heavy weight of al-Muqtabas, this means that we can only map 16,91 per cent of the entire network of articles by looking at available bibliographic information alone. {==More than four fifths are hidden from our view.==} We can currently identify only a total of 319 named authors: 139 for al-Muqtabas, 103 for al-Haqāʾiq, 52 for Lughat al-ʿArab, and 42 for al-Ḥasnāʾ{>> [See word clouds below].³⁷<<}. Quite a significant number appear only with their initials, particularly in al-Ḥaqāʾiq, and---with the exception of al-Ḥasnāʾ---all of them were men.

The first observation, similar the network of periodicals, is that only a very small number of nodes (14 of 319) are of relative importance as measured in degree (number of edges connecting to a node) and weight of the edges. In the above network plot, edges were drawn between authors when they published in the same periodical. The thickness of the edges and size of nodes are functions of the number of articles carrying the byline of a given author. Colours signify the out-degree or the number of journals an author is connected to. Authors who published only in a single journal form dense clusters{-- around that periodical's node--}. These are: al-Ḥaqāʾiq to the left, al-Muqtabas top centre, Lughat al-ʿArab bottom centre, and al-Ḥasnāʾ to the right.

If we look at the central nodes of the network more closely, we can observe that only one author published in all four journals: Maʿrūf al-Ruṣāfī was a famous poet from Baghdad, who mostly authored qaṣāʾid on current political affairs. He moved to Istanbul after the Young Turk Revolution, where he worked as an Arabic teacher at the Royal College and at the newspaper Sabīl al-Rashad. He was elected MP for al-Muthanna (Iraq) in 1912 and 1914. After WWI he became a member of the Arab Scientific Academy, established by Muḥammad Kurd ʿAlī in Damascus.³⁸ al-Ruṣāfī's close ties to al-Muqtabas and Kurd ʿAlī are further evident in the announcement in 1910 for the publication of a first collection of his poems, in which al-Muqtabas claimed that he was known among some people as "the poet of al-Muqtabas" and---wrongly---that "more than three quarters [of the qaṣāʾid therein] had been published in this journal".³⁹

The other 13 central nodes had bylines in only two out of four journals. Only eight of the fourteen authors can be found in international authority files as aggregated in VIAF (virtual international authority file), which at least means that they have not authored works catalogued in any of the contributing libraries (which, unsurprisingly, have a bias towards the Global North).⁴⁰ Those for whom we have biographic information (employing more traditional close reading of Arabic prospographic literature)⁴¹ were on average in their mid-thirties during the years under investigation. There is a {==surprising==}{>>why?<<} number of Iraqis and a notable absence of Syrians from this network of two Damascene journals and one periodical from Beirut and Baghdad each. Among the eleven identifiable authors, there are six Iraqis: Maʿrūf al-Ruṣāfī, Kāẓim al-Dujaylī, Ibrahīm Ḥilmī al-ʿAmr, Anastās Mārī al-Karmalī (often writing under the pen name Sātisnā), the two al-Shabībī brothers, Muḥammad Riḍā and Muḥammad Bāqir; three Egyptians: Muṣṭafā Ṣādiq al-Rāfiʿī, Aḥmad Muḥarram and Walī al-Dīn Yakan; and only two Syrians ʿĪsā Iskandar al-Maʿlūf and Muḥammad Rāghib Ṭabbākh. {==Less surprising is the religious composition: only two of the core nodes were Christians.==}{>>do I need this? Well there is an argument as to the dominance of Christian Arabs in the nahḍa<<} In terms of education and occupations the core nodes are exemplary for the bourgeois middle-class intelligentsia of their time: many attended Ottoman state schools in addition to more traditional, religious venues of education; many knew foreign languages in addition to Arabic and Ottoman; some were trained or even taught abroad in the colonial centres of Paris and London; some served in the Ottoman bureaucracy; some were educators. There is also a significant number of poets (7) among the central nodes⁴² and a small number of politicians (MPs). The more prolific of them were themselves journalists who at one time or another operated their own periodical(s): Maʿrūf al-Ruṣāfī, ʿIsā Iskandar al-Maʿlūf, Ibrāhīm Ḥilmī al-ʿAmr, Muḥammad Bāqir al-Shabībī, Kāẓim al-Dujaylī and Anastās Mārī al-Karmalī, the publishers of Lughat al-ʿArab. Looking at the latter, the importance of al-Muqtabas in this small network (and beyond) cannot be overstated: al-Karmalī signed more articles in al-Muqtabas than his own journal Lughat al-ʿArab. The connection to Damascus and Muḥammad Kurd ʿAlī is further evident in four of the central nodes becoming members of the Arab Scientific Academy in Damascus after World War I, whose first president was Kurd ʿAlī himself.⁴³

Another striking observation is that despite two of the journals in our corpus, al-Muqtabas and al-Ḥaqāʾiq, were published in the same city, there is only very limited overlap. This contradicts the {==common==}{>>needs reference<<} assumption that due to the very small size of local journalistic circles---in 1912, five monthly journals were published in Damascus and none had a print run of more than a few hundred copies⁴⁴---there would be a substantial overlap in authorship.

individual periodicals

The work on compiling the biographies of all 319 contributors is far from being done, but after looking at the most productive authors for each journal, we can identify certain trends in the author populations.

al-Muqtabas

Only 50 authors published more than one article in al-Muqtabas. Two of the four most prolific authors with more than ten bylines to their names wrote from Baghdad (See table below): Maʿrūf al-Ruṣāfī (24 articles) and Anastās Mārī al-Karmalī, using the pen name Sātisnā (14). ʿĪsā Iskandar al-Maʿlūf (20) wrote mostly from Zaḥle and Yūsuf Jirjis Zakham (13) from Omaha and Lincoln, Nebraska, USA. Only the fifth most prolific author was a native resident of Damascus: Muḥammad Kurd ʿAlī himself (12). {>>add comments on genre of texts by these authors<<}

The four men out of the five, for which we can find biographical records, are in many aspects exemplary of the modernising late Ottoman Empire and the Middle East: Coming from a plurality of religious and social backgrounds---Greek Orthodox, Catholic and Sunnī Muslim, priest and leading Salafi thinker of the second generation, part-time officials, of simple means and members of the old elites---they belonged to the same generation (born between the mid-1860s and mid-1870s) and worked as journalists, teachers, and occasionally politicians. All of them were highly mobile and well-travelled and had good command of local as well as foreign languages---to the extent that some of them published translations. The fifth man is not less exemplary, even though his story seems to be rather uncommon among journalists: Yūsuf Jirjis Zakham was one of the many emigrants from Greater Syria to America. He arrived in the USA in 1902 and was naturalised in 1904, settled in Lincoln, Nebraska, where he married Myra from Iowa and had at least five children. Both spouses were literate and Joseph George Zakem provided his profession as newspaper correspondent in the 1910 US Federal Census.⁴⁵

The map of relative frequencies of locations mentioned in bylines conveys the same image as the network of referenced periodicals and the brief comments on the most prolific authors' biographies: al-Muqtabas was a publication of at least regional importance. It reached well beyond Greater Syria to Egypt, Iraq and even America, turning the famous proverb "Cairo writes, Beirut publishes and Baghdad reads" upside down with Baghdad well ahead of even Damascus.

al-Ḥaqāʾiq

The picture is different for al-Ḥaqāʾiq, which was repeatedly in conflict with al-Muqtabas over the latter's supposed moral laxity. Its most prolific contributors were Damascene Sunni religious scholars from notable families, many of whom were at least one generation older than its opponents (the average year of birth for al-Ḥaqāʾiq is 1837 and 1869 for al-Muqtabas). Among them are Ibrāhīm Mardam Bek, Muḥammad ʿĀrif al-Munīr al-Ḥusaynī (b.1847/48), Mukhtār al-Muʾayyad (b.1822) and Muḥammad al-Qāsimī (b.1843), whose son Jamāl al-Dīn al-Qāsimī was among al-Muqtabas' contributors. The initially surprising finding of very limited overlap between the two networks of authors published in journals from the same city, becomes less so against this backdrop. Looking at the top 14 contributors to both journals, we can also note that whereas only two authors from al-Muqtabas are missing from VIAF and thus international library catalogues, the same is true for eight of al-Ḥaqāʾiq's most frequent authors.{>>This further affirms the difference between a journal with transregional impact and a more parochial periodical<<}

A map of the relative frequency of locations mentioned in bylines confirms the brief overview of the authors' biographies---al-Ḥasnāʾ was a parochial paper with a focus on local issues. Its geographic network was mainly restricted to Damascus itself and the cities of the Syrian hinterland.

Lughat al-ʿArab

{==missing paragraph==}{>>comment on the most prolific authors in Lughat al-ʿArab<<}

There is only limited significance in a map of relative frequency of locations mentioned in bylines for Lughat al-ʿArab. Only 26 of 939 articles in our corpus provided a location.

{====}{>>remove this map<<}

Stylometry

The final part of this paper is dedicated to the gap in authorship attribution. What are the potential means of casting some light on the more than four fifths of articles without authorship information? One approach to follow would be stylometric analysis. Stylometry is one of the methods frequently referred to as distant reading.⁴⁶ It can be summarised as the statistical analysis of literary style for the purpose of authorship attribution and genre detection, whereby "style" commonly means a frequency count of vocabulary used in a given text.⁴⁷ Stylometry is based on the empirical observation "that authors tend to write in relatively consistent, recognizable and unique ways"⁴⁸ which is particularly true for an author's vocabulary. Stylometry then computes degrees of similarity between texts, called distance measure, through comparing multivariant frequency lists of textual features. The important catch is that in order to establish similarities one has to have access to a significant corpus of digital texts by authors likely to be found among the unattributed texts. If we only compare every article in our periodical corpus to every other article in the same corpus, we cannot possibly identify any author not yet named in a byline. Instead, the best we could hope for would be to establish groups of texts that have a certain likelihood of having been authored by the same person.

There is some debate in stylometry as to which style-markers and distance measure should be considered for authorship attribution, but I settled on lists of Most Frequent Words (MFWs) and Burrow's Delta, which can be visualised as dendrograms.⁴⁹ Unlike other forms of computational linguistics, texts should not be pre-processed by, for instance, morphologising or lemmatizing. All stylometric analysis was done using the "stylo" package for R⁵⁰ on plain-text exports from the TEI files.⁵¹ Results were then visualised using Gephi.⁵²

Maciej Eder, inspired by phylogenetics, suggested to use bootstrap consensus trees and consensus networks in order to separate signal and noise and in order to overcome selection bias when picking from a range of dendrograms. In the latter method, one computes the nearest neighbour as well as the first two runners-up for a sequence of MFWs, let's say from 100 to 1000 MFWs in increments of 100, and then combine the results in a single output, which serves as a form of self-validation for the more robust signals. The results can then be visualised using {==social network analysis.==}{>>correct term?<<}⁵³ There has been some debate about the minimal required length for attribution. Eder ran a series of stylometric analyses on medium-sized corpora of prose texts in English, Polish, German, Hungarian, Latin and Greek using Burrow's Delta on 200 MFWs in order to experimentally establish a threshold length. He found that 5000 words is the minimal length for meaningful attribution below which the signal is "immensely affected by random noise".⁵⁴ Yet, contrary to common assumptions that longer text will lead to ever improving attribution results, Eder also established that beyond a length of 15000 words accuracy of authorship attribution does not improve any further. These findings have severe implications for the application of stylometry to periodicals---most texts are much shorter than 5000 words and even the longer ones are too short for random sampling. Nevertheless, our first experiments with stylometric analysis yielded promising results and show at least three distinct signals: genre, author and translator/editor.

first results

stylometry confirms what we already know

Our initial submission of a corpus of all articles longer than 3000 words to stylometric analysis using bootstrap consensus trees for 100--1000 MFWs confirmed the methods' usefulness by establishing clusters of: a) texts from the same authors or translators; b) serialised texts with or without explicit authorship information; c) texts within the same genre, such as travelogues or qaṣāʾid; d) manuscripts by different authors but with the same editor.

looking a sections of shorter articles

Many journals published shorter articles based on other publications, announcing publications etc. in sections. These articles are far {==too short ==}{>>how short, exactly? On average?<<} for reliable stylometric authorship attribution. However, we can analyse these articles as originally aggregated in sections to test the hypothesis that they were all authored by the journals' editors. We aggregated all articles in sections on the issue level, submitted them to stylometric analysis and looked for clustering by periodical. This workaround is necessary since we neither know all actual editors working at a journal, nor do we have corpora of digital texts for all known editors. Otherwise these could be directly compared to the sections.

The bootstrap consensus network of 100-1000 MWFs clearly shows clustering along periodicals based on the weight of edges (i.e. similarity) connecting sections. In the plot below al-Muqtabas is on the left, al-Ḥaqāʾiq in the centre and Lughat al-ʿArab to the right. This, at least shows that, as far as articles in sections were concerned, each periodical is stylistically distinguishable from the others.

{>>modularity<<}

Conclusion

{>>missing conclusion<<}