Zum Inhalt springen

Harnessing the Power of Text Mining in the History of Religion

Panel Chair: Edward Slingerland | Tuesday, August 25, 9-11 a.m. | Venue

Textual data generated, maintained, and transmitted by religious groups have always been central to the history of religion. The prototypical approach to textual data is a combination of qualitative methods and human synthesis, that is, we apply close readings, contextualization and theoretically motivated arguments with the purpose of interpreting those data. But with recent advances in data science, the study of religion at large is seeing new studies emerge that apply text mining methods to religious text databases. Because these studies are computationally intensive, quantitative and explanatory in scope, several methodological questions are immanent: How does text mining influence our representation of religious traditions? Can it add a qualitatively different or just a quantitatively more efficient layer to the interpretation of religious texts? To answer this, the panel will present several text mining projects and discuss the scope, status and future of text mining within the history of religion.

Logan Carson

Topic Modeling the Ancient Chinese Corpus

Our dataset is composed of 96 texts in the original language dating from the Warring States period through the Han Dynasty and beyond. Here we present and interpret topic models generated from this corpus. Topic modeling a corpus produces results in the form of clusters of words that reliably travel together through texts by a machine-learning process, and so offers an unsupervised source of information about semantic content. First, we survey the contents and proportional representations in the corpus of topic models with religious content. Second, we explore differences in religious content across Ancient China's three major philosophical traditions—Confucianism, Legalism and Daoism, with special attention to representations of high gods as opposed to mysticism. Third, we zoom in on the over 20 Confucian texts to discuss whether and how topic model results confirm or challenge conventional interpretations having to do with Confucianism and religion.

Katrine Frøkjær Baunvig

Good or bad world: Data mining the Grundtvig.dk database

To study semantic variants of the concept ‘world’ in N.F.S. Grundtvig’s (1783-1872) collected writings, we employed a suite of data mining techniques on the Danish Grundtvig database (Grundtvig.dk). N.F.S. Grundtvig was an important figure in the Danish 19th century nation building process – in shaping what was to become a modern democracy as well as in promoting the bottom-up structured churchly institution the ‘folk church’ that still unquestionably is part of the Danish cultural climate. These two ‘legacies’ reflect in Grundtvig’s authorship, which correspondingly divides in two separate domains. We found significant variation in the meaning and connotations of the concept ‘world’ across the two domains, which are likely to reflect the Christian ambivalence toward the physical ‘world’ and the liberal democratic positive evaluation. This is the conclusion of the study of two thematically different corpora from the database of Grundtvig’s writings.

Justin Lane

Semantic Networks and Texts: Analysis and Classification

Textual and linguistic analysis has been an integral part of religious studies since its inception. New computational techniques have greatly increased the efficiency of text analysis as well as our ability to quantify text data. Such techniques also open up possibilities for statistical testing. These analytical methods combine to open up new horizons in text analysis. This presentation specifically addresses how computational analysis can create more accurate, statistically based, understandings of text at the level of an individual text or corpus. The presentation defends the position that a network based approach to textual analysis allows for both the broad strokes of a corpus as well as the individuality of a text to be simultaneously represented. It also provides examples of how new statistical techniques can help support or refute earlier scholarship completed by historians. The examples drawn will come from the New Testament, Old Testament, and a multi-denominational corpus of sermons drawn from contemporary American religious congregations.

Kristoffer L Nielbo

For Allāh or kin? – Article-by-article macro-analysis of AQAP’s Inspire

As C. Geertz, among others, has argued, religious and supernatural semantics do not only function as representations of the world, but also as cultural triggers and motivators for action (Geertz 1973). Recent cognitive and evolutionary theories do, however, question the motivational strength that supernatural concepts offer when believers have to perform acts of extreme self-sacrifice. Instead they argue for a kinship semantic in which concepts related to biological ties and common ancestry are superior motivators (Atran 2010). To investigate these theoretical claims at the level of discourse, we constructed a full text database of AQAP’s (i.e., al-Qaeda in the Arabian Peninsula) Inspire magazine and modeled it using hierarchical clustering and topic modeling. Inspire is written in English and known for its combination of militant Islamist ideology and present-day digital themes (e.g., ‘open source jihad’). Results indicate that religious and kinship semantics can simultaneously compete and collaborate in the construction of a radical discursive space. This space, we argue, can induce motivational priors that facilitate concrete militant action.


B  C  D 
E  F  G  H 
I  J  K  L 
M  N  O  P 
Q  R  T 
U      V      W     XYZ 


A  B  C  D 
E  F  G  H 
I  J  K  L 
M  N  O  P 
Q  R  S  T 
U      V      W     XYZ 


Open Sessions

Thematic Outline

University Map (pdf, 192 KB)