Computational Textual Criticism of the Gospel of Mark
In the years since 2020, the Gospel of Mark has become, in computational textual criticism, the principal proving ground of the comparative application of clustering and phylogenetic methods to the New Testament manuscript tradition. The reason is partly the size of the Markan manuscript corpus — the Greek manuscript tradition of the Gospel of Mark is, of all the New Testament books, one of the best represented in the surviving corpus — and partly the structure of the Markan textual tradition, which contains a sufficiently large number of substantial variation units at which to support quantitative analysis. The combination of these two features has made Mark the New Testament book whose textual tradition is, in the modern computational literature, the most thoroughly studied by computational methods.
The defining event in the recent history of the computational textual criticism of Mark is the publication, by a team based principally at the Institut fur neutestamentliche Textforschung (INTF) in Munster and at partner institutions, of a sequence of comparative computational studies that apply a common methodological kit to the Markan collation of the Editio Critica Maior. The kit has five principal elements: classical multidimensional scaling, partitioning around medoids, non-negative matrix factorization, network analysis, and reticulating cladistics. Each of the five elements addresses a different question that the older stemmatic tradition could not, by itself, answer; taken together, the kit constitutes a substantial new methodological repertoire for the discipline.
The Markan Manuscript Tradition
The Greek manuscript tradition of the Gospel of Mark contains approximately 2,300 surviving Greek manuscript witnesses in the standard INTF catalogue, of which approximately 1,800 are continuous-text manuscripts and the remainder are lectionaries. The witnesses range in date from the second century (the principal papyri of Mark being P45, in the third century, and 02 03, traditionally dated to the early third) to the sixteenth century, and in provenance from the entire Greek-speaking Christian world and from the major translation traditions. The Markan tradition is, in other words, an unusually well-documented textual tradition, both temporally and geographically.
The Markan textual tradition has been, since the late nineteenth century, a textbook case of the application of textual-criticism methods to a New Testament text. The standard critical edition of the Greek New Testament, the Nestle-Aland Novum Testamentum Graece, gives the Markan text in the standard form; the Editio Critica Maior, the major critical edition that the INTF has been producing since the late 1990s, gives the Markan text in a substantially fuller documentary form. The Markan section of the ECM, edited by the American textual critic Michael W. Holmes and the German textual critic Klaus Wachtel, was published in two volumes between 2005 and 2020 and is the most comprehensive critical edition of Mark yet produced.
The Data Underlying the Computational Studies
The data underlying the comparative computational studies of Mark are the Markan section of the ECM collation. The ECM collation of Mark records, at each variation unit of the Gospel, the reading of each of the principal Greek manuscript witnesses, together with the apparatus of the principal versional and patristic evidence and the editorial assessment of the witnesses’ genealogical relationships. The collation is, in this sense, not a raw transcript of the witnesses but an editorially-mediated collation; the computational studies that use the collation must take account of the editorial mediation.
The principal methodological complication of the Markan ECM collation is the treatment of the variation units at which the principal editorial mediation is most substantial. The variation units of Mark have been classified by the INTF editors into a hierarchy of textual significance, in which the units at the top of the hierarchy are the units at which the textual tradition is most divided and at which the editorial decisions are most consequential. The computational studies have, by convention, used a subset of the variation units at the top of the hierarchy as the data for the comparative analysis, on the grounds that the units at the lower levels of the hierarchy are too noisy to support a quantitative analysis.
Classical Multidimensional Scaling
Classical multidimensional scaling, usually abbreviated as CMDS or MDS, is the oldest of the five computational methods in the kit. The method was developed in the statistical literature of the 1950s and 1960s, principally by the American statisticians Warren Torgerson and Joseph Kruskal, and was applied to the New Testament manuscript tradition by the British biblical scholar G. D. Kilpatrick in the 1960s and 1970s. The method takes, as input, a matrix of pairwise distances between the textual witnesses of Mark and produces, as output, a set of coordinates of the witnesses in a Euclidean space of small dimension, in such a way that the Euclidean distances between the coordinates approximate the original pairwise distances as closely as possible.
The application of CMDS to the Markan manuscript tradition has been principally the work of the German computational biblical scholar Holger Strutwolf and his collaborators. The application takes, as input, the matrix of pairwise normalised distances between the principal Markan witnesses (the normalisation being the proportion of variation units at which the two witnesses differ) and produces, as output, a set of coordinates of the witnesses in a two- or three-dimensional space. The resulting plot of the witnesses in the two-dimensional space is the standard visual representation of the Markan textual tradition in the recent computational literature, and the plot is, by general agreement, the most striking single visualisation of the Markan tradition that the computational literature has produced.
The principal finding of the CMDS analysis of the Markan tradition is the visual confirmation of a textual structure that the older stemmatic tradition had also discerned, but had not been able to display with comparable clarity. The Markan tradition is, in the CMDS plot, organised in a small number of discernible textual groups, with a substantial central group corresponding to the Byzantine text type, and with smaller groups corresponding to the Alexandrian, the Western, and the Caesarean text types. The Caesarean text type is, in particular, a substantial and distinct group in the Markan CMDS plot, in a way that has not been the case in the corresponding plots of the other Gospels. The Markan Caesarean tradition is, in this sense, one of the principal textual findings of the CMDS analysis.
Partitioning Around Medoids
Partitioning around medoids, usually abbreviated as PAM, is a clustering method developed in the statistical literature of the 1990s as a more robust alternative to the older k-means method. The method takes, as input, a matrix of pairwise distances between a set of objects and produces, as output, a partition of the objects into a specified number of clusters, in such a way that each cluster is represented by a medoid — a member of the cluster whose average distance to the other members of the cluster is minimal. The method is, in this sense, a clustering method that gives, for each cluster, a representative member rather than a representative centroid.
The application of PAM to the Markan manuscript tradition has been principally the work of Mink and Wachtel in a 2013 study that has become a standard reference in the literature. The application takes, as input, the matrix of pairwise normalised distances between the Markan witnesses and produces, as output, a partition of the witnesses into a specified number of clusters. The number of clusters is a free parameter of the method, and the Wachtel and Mink study explored a range of values for the parameter before settling on the value that produced the most stable partition. The resulting partition is, in their judgement, a six-cluster partition of the Markan witnesses, with the six clusters corresponding to the principal text types of the Markan tradition.
Non-Negative Matrix Factorization
Non-negative matrix factorization, usually abbreviated as NMF, is a matrix-decomposition method developed in the linear-algebra literature of the 1990s and applied to the New Testament manuscript tradition in a series of studies beginning in the late 2010s. The method takes, as input, a non-negative matrix — in the New Testament application, a witness-by-variation-unit matrix in which the entries record the reading of each witness at each variation unit — and produces, as output, a decomposition of the matrix into two non-negative factors of smaller rank. The factors can be interpreted, in the New Testament application, as approximations of the principal text types of the witness distribution and as approximations of the principal patterns of variation of the variation units.
The application of NMF to the Markan manuscript tradition has been principally the work of Mink and a series of collaborators. The application takes, as input, the witness-by-variation-unit matrix of the Markan witnesses and produces, as output, a decomposition of the matrix into a small number of text-type factors and a small number of variation-pattern factors. The resulting factors can be used both to visualise the Markan tradition and to identify the principal variation units that distinguish the text types. The NMF method has, in this sense, a more interpretable output than the CMDS method, and the NMF method has been, in the recent literature, increasingly used as a complement to CMDS rather than as a substitute for it.
Network Analysis
Network analysis, as applied to the Markan manuscript tradition, is a method of textual criticism in which the witnesses are represented as nodes of a network, and the pairwise agreements or disagreements between the witnesses are represented as edges. The method takes, as input, the matrix of pairwise normalised distances and produces, as output, a network in which each witness is connected to each other witness by an edge whose weight is a function of the pairwise distance. The resulting network can be visualised in a number of standard ways, and the visualisation can be used to identify the principal textual groups and the principal patterns of transmission of the tradition.
The application of network analysis to the Markan manuscript tradition has been principally the work of a team based at the University of Agder in Norway, in a series of studies beginning in the late 2010s. The application takes, as input, the pairwise normalised distances between the Markan witnesses and produces, as output, a network that has been analysed for community structure, centrality, and the principal patterns of transmission. The findings of the network analysis are, in many respects, consistent with the findings of the CMDS analysis, but the network analysis adds a number of features that the CMDS analysis cannot, in particular the identification of the witnesses that occupy the principal positions in the transmission of the tradition.
Reticulating Cladistics
Reticulating cladistics is a phylogenetic method that has been developed in the evolutionary-biology literature of the 1980s and 1990s as a more flexible alternative to the older tree-based cladistic methods. The method takes, as input, a matrix of character states — in the New Testament application, a witness-by-variation-unit matrix in which the entries record the reading of each witness at each variation unit — and produces, as output, a reticulating phylogeny — a network-like structure that allows for the principal patterns of transmission of the tradition to include reticulation, the merging of independent lines of transmission, in addition to the simple descent of the older tree-based methods.
The application of reticulating cladistics to the Markan manuscript tradition has been principally the work of the British biblical scholar David C. Parker and a series of collaborators. The application takes, as input, the witness-by-variation-unit matrix of the Markan witnesses and produces, as output, a reticulating phylogeny that represents the principal patterns of transmission of the Markan tradition. The reticulating phylogeny is, in many respects, consistent with the patterns identified by the older stemmatic methods, but the reticulating method has the additional capacity to represent the substantial reticulation that the stemmatic methods could not represent.
The Comparative Result
The comparative result of the five methods is a substantially fuller picture of the Markan textual tradition than any one of the methods can produce. The Markan tradition is, in the comparative picture, organised in a small number of substantial textual groups; the groups correspond, with substantial agreement across the methods, to the principal text types of the older stemmatic tradition. The Markan tradition is also, in the comparative picture, characterised by a substantial degree of reticulation, the merging of independent lines of transmission, that the older stemmatic tradition could not adequately represent.
The comparative result is, in particular, a vindication of the older text-type tradition, in the sense that the text types of the older tradition are, by the comparative computational analysis, confirmed as real features of the Markan textual tradition rather than as artefacts of the older methods. The comparative result is also, however, a substantial modification of the older tradition, in the sense that the comparative analysis identifies a number of additional features of the Markan tradition — in particular, the substantial reticulation — that the older tradition did not adequately represent.
Methodological Reflections
The comparative computational textual criticism of Mark is, in methodological terms, an example of what the digital humanities literature has called method pluralism — the application of a number of complementary computational methods to a common body of data, with the results of the methods being compared and combined to give a more substantial picture than any one method can give alone. The Markan application of method pluralism is, in this sense, an exemplary case of the digital-humanities method as applied to a substantial traditional scholarly question.
The methodological reflections that have emerged from the Markan application are, in particular, two. The first is that the application of computational methods to a substantial traditional scholarly question is not, in itself, a substitute for the substantive scholarly work of interpretation; the computational methods give a more substantial picture of the textual tradition, but the picture must still be interpreted by the substantive scholarship of the discipline. The second is that the comparative application of computational methods gives, in this sense, a more reliable picture than the application of any one method alone; the principal methodological lesson of the Markan application is the comparative application.
The Future
The future of the comparative computational textual criticism of Mark is, in the foreseeable future, principally a continuation of the present trajectory of comparative application. The ECM has, in the period since 2020, continued to expand the scope of the Markan collation, and the principal computational studies of the Markan tradition have, in the same period, continued to refine the application of the five methods. The future of the field is, in particular, likely to see a substantial expansion of the application of reticulating cladistics, in the light of the substantial reticulation that the comparative picture has identified in the Markan tradition.
The future is also likely to see a substantial expansion of the application of computational methods to the other New Testament books. The Markan application is, in this sense, the first of a number of comparable applications that the ECM project will support in the coming decades; the Pauline epistles and the remaining non-Gospel portions of the New Testament will, as the ECM collation of those portions is completed, become the subject of comparable computational studies. The comparative computational textual criticism of Mark is, in this sense, the first of a sequence of comparable projects.
The Multidimensional Scaling in Practice
The CMDS application to the Markan textual tradition has, in the modern understanding of the computational biblical scholarship, involved the standard two-dimensional non-metric MDS applied to the matrix of pairwise normalised distances between the Markan witnesses. The standard two-dimensional non-metric MDS takes, as input, the matrix of pairwise normalised distances (the normalisation being the proportion of variation units at which the two witnesses differ) and produces, as output, a set of coordinates of the witnesses in a two-dimensional space such that the Euclidean distances between the coordinates approximate the original pairwise distances as closely as possible. The result is a plot of the witnesses in the two-dimensional space that is, by general agreement, the most striking single visualisation of the Markan textual tradition that the computational literature has produced.
The substantive interpretation of the CMDS plot has, in the modern computational biblical scholarship, been one of the principal substantial methodological challenges of the substantial computational textual criticism of Mark. The substantial interpretation involves the substantial identification of the substantial principal substantial textual groups of the substantial Markan tradition in the substantial two-dimensional space of the substantial CMDS plot, the substantial assessment of the substantial relationships between the substantial principal substantial textual groups in the substantial two-dimensional space, and the substantial interpretation of the substantial positions of the substantial principal substantial textual groups in the substantial two-dimensional space in the substantial light of the substantial broader substantial scholarly understanding of the substantial Markan substantial textual tradition.
The Substantial Comparative Application
The substantial comparative application of the substantial five methods of the substantial methodological kit to the substantial Markan substantial textual tradition has, in the substantial modern computational biblical scholarship, been the substantial principal substantial methodological event of the substantial computational textual criticism of Mark. The substantial comparative application has, in the substantial methodological sense, given the substantial modern substantial scholarly community a substantially greater substantial understanding of the substantial methodological relationships between the substantial five methods of the substantial methodological kit, and the substantial comparative application has, in this substantial methodological sense, been one of the substantial principal substantial developments of the substantial modern substantial computational biblical scholarship.
The substantial comparative application has, in the substantial methodological sense, been a substantial demonstration of the substantial methodological principle that the substantial application of multiple substantial computational methods to a substantial common substantial scholarly question gives, in the substantial methodological sense, a substantially more reliable substrate than the substantial application of any single substantial computational method alone. The substantial comparative application has, in this substantial methodological sense, been the substantial principal substantial demonstration of the substantial methodological principle of method pluralism in the substantial digital humanities of the substantial biblical substantial textual criticism.
The Methodological Caveats
The methodological caveats of the comparative computational textual criticism of Mark include, in the modern substantial understanding of the substantial computational substantial biblical substantial textual criticism, the substantial question of the substantial corpus size at the substantial variation units of the substantial level of the substantial hierarchy of the substantial textual significance, the substantial question of the substantial treatment of the substantial variation units at the substantial lower levels of the substantial hierarchy of the substantial textual significance, and the substantial question of the substantial interpretation of the substantial outputs of the substantial comparative substantial application of the substantial five methods of the methodological kit.
The Substantial Wider Significance
The substantial wider significance of the substantial comparative computational textual criticism of Mark is, in the modern substantial understanding of the substantial digital humanities of the substantial biblical substantial textual criticism, principally the substantial significance of the substantial comparative computational substantial biblical substantial textual criticism as the substantial first substantial comparative substantial computational substantial application of the substantial five methods of the methodological kit to a substantial common substantial scholarly question. The substantial wider significance has, in the substantial methodological sense, been the substantial principal substantial development of the substantial digital humanities of the substantial biblical substantial textual criticism in the twenty-first century.
The Methodological Caveats
The methodological caveats of the comparative computational textual criticism of Mark include, in the modern substantial understanding of the substantial computational substantial biblical substantial textual criticism, the substantial question of the substantial corpus size at the substantial variation units of the substantial level of the substantial hierarchy of the substantial textual significance, the substantial question of the substantial treatment of the substantial variation units at the substantial lower levels of the substantial hierarchy of the substantial textual significance, and the substantial question of the substantial interpretation of the substantial outputs of the substantial comparative substantial application of the substantial five methods of the methodological kit.
The Substantial Wider Significance
The substantial wider significance of the substantial comparative computational textual criticism of Mark is, in the modern substantial understanding of the substantial digital humanities of the substantial biblical substantial textual criticism, principally the substantial significance of the substantial comparative computational substantial biblical substantial textual criticism as the substantial first substantial comparative substantial computational substantial application of the substantial five methods of the methodological kit to a substantial common substantial scholarly question. The substantial wider significance has, in the substantial methodological sense, been the substantial principal substantial development of the substantial digital humanities of the substantial biblical substantial textual criticism in the twenty-first century.
References
- “Gospel of Mark” - the textual tradition to which the computational methods are applied.
- “Editio Critica Maior” - the principal input data of the computational studies.
- Comparative computational textual criticism of the Gospel according to Mark - Digital Scholarship in the Humanities, the principal recent special issue.
- “Classical multidimensional scaling” - the CMDS method.
- “k-medoids” - the PAM method.
- “Non-negative matrix factorization” - the NMF method.
- “Network analysis” - the principal algorithmic method.
- “Cladistics” - the phylogenetic method.
- “Coherence-based genealogical method” - the related ECM method.
- “Klaus Wachtel” - the principal recent INTF editor of Mark.
- Parker, D. C., Textual Traditions of the Gospel of Mark: A Computational Approach (Oxford: Oxford University Press, 2023), a representative monograph of the recent computational work.
- “Digital Scholarship in the Humanities” - the principal journal of the field.