Since October 2018, version 1.0rc of a digital genetic edition of Goethe’s Faust is publicly accessible online (Goethe 2018). The poster visualizes and explains the edition’s most specific features and discusses recent research about modelling macrogenesis as a graph.

The archive is the edition’s heart piece with detailed multi-view representations of about 550 manuscripts spanning a period from the early 1770s until 1832. Additionally, it contains printed editions and testimonies about the genesis of Faust (references found in diaries, letters etc.). For each of the manuscripts there are the following views, some of which are combined on screen:

Additionally, each manuscript’s final textual stage is rendered in way that closely corresponds to the rendering of the critically established text of the work (see below). All but the diplomatic views are also available for the printed editions.

The edition’s genesis section comes with navigable graphic visualizations of how witnesses relate to each other. Bar diagrams (see fig. 3) indicate the textual content of each witness by length of and gaps within bars. Their order is based on a global chronological ordering expressed in a directed acyclic graph (see below for details about macrogenesis).

The edition’s text was critically established on the basis of the complete textual transmission. By clicking on a line of text all other versions appear synoptically (fig. 4), showing its development within and across all other witnesses in graph based chronological order (see below). All sigils are links that lead to the corresponding passage of the respective witness in the archive (see above). In addition to the synopsis, there are critical apparatus entries that report about editorial decisions.

The views are generated from TEI-encoded source data, the poster will illustrate the process.

The chronological (i.e. macrogenetic) order of witnesses is far from trivial: Only a few witnesses can be precisely dated. There are many relative chronologies that only span a few witnesses each. In an experimental part of the edition, all relevant chronological information has been extracted from scholarly literature (most notably Goethe, 1888; Fischer-Lamberg, 1955; Bohnenkamp, 1994) and recorded in a simple data model that basically recognizes relative and absolute datings, together with the bibliographic source and optional comments. This information has been modeled as a directed graph, as illustrated in fig. 5: E.g., the path 1825-05-25 → 2 III H.5 → 1825-04-06 means that 2 III H.5 has been written not before February 26th and not after April 4th, 1825. From this graph, additional information can be inferred, e.g., limits for the absolute dating of 2 III H.8.

Contradictions in model or sources lead to cycles in the graph (fig. 6). In the recorded information, there is one strongly connected component (in which every node is reachable from every other node) composed of 477 witnesses and 2136 assertions, too many to check manually. Identifying and removing all contradictions with the smallest number of interventions means solving the NP-complete (Karp, 1972) minimum feedback arc set problem, for which we can calculate an exact solution using a method by Baharev et al. (2015). The procedure can be manually influenced by placing weights on sources or individual edges or by marking obviously wrong edges as to ignore first.

From the cycle-free results, a topological ordering of the witnesses can be created (Manber, 1998:199) that is then used everywhere the edition requires a chronological ordering. Subgraph visualisations created using NetworkX (Hagberg et al., 2008) and GraphViz (Gansner/North, 2000) help to explain the ordering.