From Manuscripts to Archetypes through Iterative Clustering

From Manuscripts to Archetypes through Iterative Clustering Armin Hoenen CEDIFOR Goethe university Frankfurt hoenen@em.uni-frankfurt.de Abstract Corpora of manuscripts of the same ancient text often preserve many variants. This is so because upon copying over long copy chains errors and editorial changes have been repeatedly made and reverted to the effect that most often no 2 variant texts of the same so-called textual tradition have exactly the same text. Obviously in order to save the time to read all of the versions and in order to enable discourse and unambiguous referencing, philologists have since the beginnings of the age of print embarked on providing one single textual representation of this variety. In computational terms one tries to retrieve/compose the base text which is most likely the latest common ancestor (archetype) of all observed variants using stemmata – that is trees depicting the copy history (manuscripts = nodes, Copy processes = edges). Recently, they have been computed and evaluated automatically (Roos and Heikkilä, 2009). Likewise, automatic archetype reconstruction has been introduced lately (Hoenen, 2015b; Koppel et al., 2016). A synthesis of both stemma generation and archetype reconstruction has not yet been achieved. This paper therefore presents an approach where through iterative clustering a stemma and an archetype text are being reconstructed bottom-up. Keywords: stemmatology, clustering, archetype text reconstruction, digital textual criticism 1. Introduction For corpora of ancient texts transmitted (and born) in the age of handwriting, collections of those documents tend to have variant texts which differ slightly due to a variety of reasons among which miscopying, editorial changes such as modernization and stylistic changes may be the most well known due to their illustrativity. In order to provide one text version to the interested scholar, generations of philologists have tried to offer single texts (even though some of those are embedded as base texts in edition types such as critical editions) which they argue best represent the most probably lost authorial original and which some- how mimick an edicio princeps initial print exemplar of an autograph.1 Computationally, the task involves various subtasks. First, manuscript texts need to be digitized, for instance through manual retyping and possibly encoding. To this end, the Text Encoding Initiative2 provides the el- ements and rules for an appropriate schema. Texts can be manually or computationally aligned and from this various algorithms can produce stemmata, see Figure 1 that is trees with nodes representing manuscripts and edges represent- ing copy processes or chains thereof, which graphically dis- play (a hypothesis about) the copy history of the variants. Although Cameron (1987) observes that the goal of a tex- tual critic3 or philologist is rather the reconstruction of an urtext4 than of the copy history and that the copy history 1Consider Shillingsburg (2017) for discussions on (other) pur- poses of editions. 2http://www.tei-c.org/ 3Textual criticism is the philological discipline of critically re- flecting the given variation in a tradition. 4The term urtext is a loan from German which may refer to the very first (complete) version of a text. The …

Content truncated to 500 words