Catch me if you can! Tracing the late Ottoman ideosphere through network analysis and stylometry of the Arabic periodical press

Till Grallert, Orient-Institut Beirut (OIB), @tillgrallert

3rd Online Conference of the Islamicate Digital Humanities Network (IDHN)

29 April 2020

Slides: https://OpenArabicPE.github.io/slides/2020-idh/

Overview

Problematic state of Arabic periodical studies
Methodologic proposals to computationally address this state
My approaches
- bibliometrics
- (social) network analysis
- stylometric authorship attribution

Problems

periodicals are commonly perceived as a source but not a subject in its own right
the history of the periodical press
- focusses on a few “core” publications from Cairo and Beirut
- remains largely unexplored
- is full of untested hypotheses
- is heavily biased by national(ist) narratives

One of the many reasons

Map: geographic distribution of library holdings of al-Muqtabas

Open questions:

micro level

how were individual periodicals produced?
who authored the vast majority of anonymous articles?

meso level

What are the core nodes (authors, periodicals, other works) in the ideosphere of the late Ottoman Eastern Mediterranean?

macro level

How do we have to revise the nahḍa narrative / late Ottoman intellectual history if we include the full ideosphere of the press beyond Cairo and Beirut?

Methodology: computational approaches

distant reading
- bibliometrics
- social network analysis
- stylometric authorship attribution
requires a digital corpus -> corpus building

corpus building

It’s labour and resource intensive. It really is!

get the data
- text: transcription, train OCR/HTR
- facsimiles: scanning
- bibliographic metadata: transcription, validated iterative generation
transform the data into a human and machine readable edition
- model the source
- identify entities and link them to authority files
host, share and preserve the data

corpus building: Open Arabic Periodical Editions

ideas:
- unite available facsimiles and transcriptions in a standard-compliant format
- harvest, generate, validate and share open metadata
aims
- validate the transcription against the facsimiles
- improve the transcription with the help of the “crowd”
- make everything citable for scholars, linkable for machines
- share all data, metadata and tools with the broadest possible licences to facilitate access and re-use
principles
- re-purpose available and established tools, technologies, and material
- preference for open and simple formats and tools

corpus building: Open Arabic Periodical Editions

Open licences: CC BY-SA 4.0 (TEI, MODS, BibTeX), MIT license (XSLT, XQuery)
Social digital editions hosted on GitHub: gradually improve transcription and mark-up
Releases are archived at Zenodo: receive a DOI for reliable citation
Static web-view: provides side-by-side view of facsimiles and text
Access to bibliographic metadata through a public Zotero group

OpenArabicPE’s corpus

periodical	doi	volumes	issues	articles	words	words per article
al-Ḥaqāʾiq	10.5281/zenodo.1232016	3	35	389	298090	832.66
al-Ḥasnāʾ	10.5281/zenodo.3556246	1	12	201	NA	NA
al-Manār		35	537	4300	6144593	1437.73
al-Muqtabas	10.5281/zenodo.597319	9	96	2964	1981081	873.34
al-Ustādh	10.5281/zenodo.3581028	1	42	435	221447	582.21
al-Zuhūr	10.5281/zenodo.3580606	4	39	436	292333	695.09
Lughat al-ʿArab	10.5281/zenodo.3514384	3	34	939	373832	485.21
total		56	795	9664	9311376

OpenArabicPE’s corpus

title	articles with author (%)	authors	author.birth.avg	author.death.avg
al-Ḥaqāʾiq	41.90	104	1837.53	1905.93
al-Ḥasnāʾ	37.31	50	1883.14	1947.17
al-Manār	87.14	356	1870.96	1913.5
al-Muqtabas	12.72	140	1855.94	1929.48
al-Ustādh	5.52	8	NA	NA
al-Zuhūr	41.51	112	1872.96	1939.52
Lughat al-ʿArab	16.19	53	1875.8	1942.1

First attempts to computationally map the ideosphere of the late Ottoman Arabic-speaking press

Evaluating the corpus: network of referenced periodicals

network of referenced periodicals: data sources

mark-up of all references to periodicals in the text
- semi-automatic
authority files for disambiguation and additional information
- mostly automatic

والأصح الدرعية بلام التعريف (راجع <bibl subtype="journal" type="periodical">مجلة <title level="j" ref="oape:bibl:3 oclc:1034545644" xml:id="title_16.d2e2291">الزهور</title> المصرية  <biblScope unit="volume" from="2" to="2">٢</biblScope> :  <biblScope unit="page" from="292">٢٩٢</biblScope></bibl>)

وانتخب <persName>فؤاد أفندي الدفتري البغدادي</persName> و<bibl><editor><persName>نوري أفندي</persName></editor> راس كتاب <textLang otherLangs="ota">القسم التركي</textLang> في <bibl type="periodical" subtype="newspaper">جريدة <title ref="oape:bibl:532">الزهور</title></bibl> البغدادية</bibl> نائبين عن <placeName ref="oape:place:372 geon:94824">كربلاء</placeName>.

network of referenced periodicals

Figure: Network of periodicals mentioned al-Ḥaqāʾiq, al-Ḥasnāʾ, Lughat al-ʿArab and al-Muqtabas; weights per issue

only a few nodes are of relative importance (44 of 465)
al-Muqtabas accounts for the vast majority of references
all periodicals are primarly self-referential

network of referenced periodicals: the core

Figure: Core nodes in the network of periodicals mentioned al-Ḥaqāʾiq, al-Ḥasnāʾ, Lughat al-ʿArab and al-Muqtabas; weights per issue

confirms the bias on Cairo and Beirut (8 of 9)
highly centralised in terms of geographic distribution (10 locations)
surprising members

Bibliometrics: the network of authors

network of authors: data sources

structured bibliographic data provided by the text itself
- semi-automatic
- problems: many accronyms, plurality of name forms
authority files for disambiguation and additional information
automatic enriching: from semantic web
- life dates
- works
- geocoded locations

<person>
    <persName><roleName type="pseudonym">ساتسنا</roleName></persName>
    <persName><roleName type="pseudonym">أمكح</roleName></persName>
    <persName><roleName type="pseudonym">فهر الجابري</roleName></persName>
    <persName><roleName type="rank">الأب</roleName> <forename>أنستاس</forename> <forename>ماري</forename> <surname><addName type="nisbah">الكرملي</addName></surname></persName>
    <persName><forename>أنستاس</forename> <forename>ماري</forename> <addName type="nisbah">الألياوي</addName> <surname><addName type="nisbah">الكرملي</addName></surname></persName>
    <persName><forename>بطرس</forename> <addName type="nasab">بن <forename>جبرائيل</forename></addName> <forename>يوسف</forename> <surname>عواد</surname></persName>
    <idno type="VIAF">39370998</idno>
    <idno type="oape">227</idno>
    <idno type="wiki">Q4751824</idno>
    <birth><date source="viaf" when="1866-08-05">1866-08-05</date> in <placeName ref="oape:place:216 geon:98182">Baghdad</placeName></birth>
    <death><date source="viaf" when="1947-01-07">1947-01-07</date> in <placeName ref="oape:place:216 geon:98182">Baghdad</placeName></death>
</person>

network of authors

Figure: Network of authors with bylines in al-Ḥaqāʾiq, al-Ḥasnāʾ, Lughat al-ʿArab and al-Muqtabas

only a few nodes are of relative importance
limited overlap between journals from the same city
nodes are connected by various other social networks (education, imperial service, family relations etc.)

network of authors: the core

Figure: Core nodes in the network of authors with bylines in al-Ḥaqāʾiq, al-Ḥasnāʾ, Lughat al-ʿArab and al-Muqtabas

many Iraqis
few Syrians
few Christians
many poets
majority not covered by scholarly literature

network of authors: geography

Map: Locations in bylines in al-Muqtabas (Cairo and Damascus)

Map: Locations in bylines in al-Ḥaqāʾiq (Damascus)

Map: Locations in bylines in al-Ḥasnāʾ (Beirut)

Map: Locations in bylines in Lughat al-ʿArab (Baghdad)

problem: network reflects only 17% of articles

finally: computational authorship attribution

state of the field

question of authorship attribution has not received much attention
implicit and commonly accepted hypothesis: editors authored all anonymous articles themselves

problems:

hypothesis remains untested
we don’t know the potential candidates for authorship even within the hypothesis
it is unlikely that a single person authored everything

computational authorship attribution: stylometry

this has not yet been applied to Arabic
method: compare stylistic features to gain a numerical measure of difference
- distance measure is sensitive to composition of corpus
stylistic features: most frequent words (MFWs)
- have been shown to be extremely effective for authorship attribution
chunking/sampling impacts results
- features: run multiple iterations and have them vote
- texts: minimum length of 4000-5000 words

stylometry: 5000 words

Figure: bootstrap consensus network of articles (length >= 5000 words, 100–1000 MFWs), colours by modularity group

it works! correctly identified signal of
- authorship
- editorship
- translation
additional (sub)-signal: genre
contradicts the hypothesis: anonymous author who is not the editor

stylometry: how to deal with the minimal length requirement?

Figure: bootstrap consensus network of sections and articles (length >= 5000 words, 100–1000 MFWs) in al-Muqtabas, coloured by modularity

oape.878: Muḥammad Kurd ʿAlī
oape.741: Charles Seignobos (in translation by Muḥammad Kurd ʿAlī)
oape.482: Muḥammad Jamāl al-Dīn al-Qāsimī
oape.707: Aḥmad Zakī Pasha
oape.632: Jirjī Ḥaddād

stylometry: how to deal with the minimal length requirement?

Figure: bootstrap consensus network of sections of articles in 123 periodical issues based on consensus of 100–1000 MWFs, coloured by modularity

Summary

corpus building: modelling is indispensible for article-level analysis
network of periodicals: confirms importance of Cairo
- shows vast differences between journals: parochial, regional, trans-regional
- should guide corpus building
network of authors: surprising
- importance of Iraq and Iraqis
- very limited overlap between journals from the same city
- many authors not mentioned in standard accounts
stylometry: promising
- problem: 5000 word minimum length

Thank you!

Contributors to OpenArabicPE: Jasper Bernhofer, Dimitar Dragnev, Patrick Funk, Talha Güzel, Hans Magne Jaatun, Jakob Koppermann, Xaver Kretzschmar, Daniel Lloyd, Klara Mayer, Tobias Sick, Manzi Tanna-Händel and Layla Youssef
Links:
- Slides: https://OpenArabicPE.github.io/slides/2020-idh/
- Paper (pre-print, submitted): https://doi.org/10.5281/zenodo.1413610
- Project URL: https://www.github.com/OpenArabicPE
- Project blog: https://openarabicpe.github.io
- Twitter: @tillgrallert
- Email:
Licence: slides and plots are licenced as CC BY-SA 4.0

Catch me if you can! Tracing the late Ottoman ideosphere through network analysis and stylometry of the Arabic periodical press

Overview

Introduction

Problems

One of the many reasons

Open questions:

micro level

meso level

macro level

Methodology

Methodology: computational approaches

corpus building

corpus building: Open Arabic Periodical Editions

corpus building: Open Arabic Periodical Editions

OpenArabicPE’s corpus

OpenArabicPE’s corpus

First attempts to computationally map the ideosphere of the late Ottoman Arabic-speaking press

Evaluating the corpus: network of referenced periodicals

network of referenced periodicals: data sources

network of referenced periodicals

network of referenced periodicals: the core

Bibliometrics: the network of authors

network of authors: data sources

network of authors

network of authors: the core

network of authors: geography

problem: network reflects only 17% of articles

finally: computational authorship attribution

state of the field

problems:

computational authorship attribution: stylometry

stylometry: 5000 words

stylometry: how to deal with the minimal length requirement?

stylometry: how to deal with the minimal length requirement?

Conclusion

Summary

Thank you!