Abstract
As with other pre-modern Islamic movements, a significant temporal gap separates the hypothesized origins of Sufism in the mid-9th century from the establishment of an extensive documentary record in the late 10th c. This situation raises equally significant historiographical questions as to whether later Sufi literature reflects the biases of later generations rather than genuinely early doctrines (Mojaddedi 2001; Knysh 2017; Salamah-Qudsi 2018). This paper proposes a method to analyze the transmission of early Sufi doctrines by applying computational techniques of textual analysis to a corpus of the teachings of the leading figure of Baghdadi Sufism, Abu al-Qasim al-Junayd (d. 910–11).
The record of Junayd's teachings provides a case study of the difficulties associated with the early Sufi textual tradition. Whereas the contents of seven of his treatises (ed. 1969) advance bold mystical theses and evince a highly abstract style, the hundreds of sayings ascribed to him in biographical works (Sulami 1953; Abu Nu'aym 1932–38) often depict a more cautious, "sober" epistemology and mundane phraseology. Moreover, several scholars have noted that doctrines expressed in both the epistles and late biographical literature are clearly fabricated (Arberry 1953; Radtke 2005). Absent the discovery of new textual evidence, insight into the history of Junayd's doctrines must be gained through novel means.
Topic modeling——the analysis of the distribution of lexical and syntactic patterns within and across groups of texts——provides the requisite framework for this inquiry. Theoretically, if the corpus of Junayd's doctrines genuinely originated with him, the results of stylometric analysis would reflect uniform distributions of textual patterns; conversely, individual alterations to Junayd's doctrines would yield notably differing distributions.
In order to evaluate the corpus of Junayd's doctrines, I first partition it into individual textual units including: (1) each of the aforementioned treatises; or (2) the teachings in the biographical works transmitted along a uniform "citation pathway" (Ar. isnad). A preliminary assessment of this latter group indicates that nearly 80% of the teachings attributed to Junayd were transmitted through 10 such isnads. Methodologically, after preprocessing a digital edition of the relevant text(s) for each unit, I train a series of topic models through Bayesian Optimization and create a matrix of similarity scores using the Octis package for Python (Terragni 2022). The proposed research not only sheds new light on the development of early Sufi doctrines, but contributes to the growing field of computational analysis of premodern Arabic-language texts.
Discipline
Geographic Area
Sub Area
None