Critical Digital Humanities: The Search for a Methodology

Critical Digital Humanities: The Search for a Methodology
by James E. Dobson

The University of Illinois Press, Urban-Champaign, 2019
196 pp., illus. 11 b/w. Trade, $99; paper, $25; ebook, $19.95
ISBN: 978-0-252-04227-0; ISBN: 978-0-252-08404-1; ISBN: 978-0-252-05111-1.

Reviewed by
Jan Baetens
January 2020

Earlier this year Nan Z. Da, teacher of literature at Notre-Dame, published an article on the discrepancy between computational analysis and literary studies that opens with a smashing claim: “In a nutshell the problem with computational literary analysis as it stands is that what is robust is obvious (in the empirical sense) and what is not obvious is not robust, a situation not easily overcome given the nature of literary data and the nature of statistical inquiry. There is a fundamental mismatch between the statistical tools that are used and the objects to which they are applied” (“The Computational Case against Computational Literary Studies,” Critical Inquiry Spring 2019, vol. 45, number 3, p. 601). At the time this article appeared, James E. Dobson’s book had already been printed (it was released in March), if not the author would certainly have added this and other quotations by the same author to his work, which is exactly about that: the problems that arise when doing Digital Humanities, the very first and perhaps also last problem being the uncritical belief in the endless benefits and possibilities of applying computational techniques to humanities. This is not to say that Dobson’s book has the same scope as Da’s essay: The former does not limit its research to the field of literary studies alone (Dobson is interested in the humanities, which includes for instance history––but not linguistics; he also explicitly leaves aside the social sciences), while it also relies on a different, more historical and philosophical way of shaping its fundamental suspicion of statistical and logarithmic tools and perspectives in humanities research.

As clearly stated in the book’s subtitle, this is not a study against Digital Humanities, which Dobson considers an important and necessary innovation that asks radically new questions and offers a wide range of new opportunities. It is instead a reflection in favor of a critical Digital Humanities, that is a specifically humanist take on computational analysis that Dobson considers missing in much Digital Humanities research and that he fundamentally links with the notion of suspicion. The global theoretical framework of the book is thus strongly indebted to the spirit of deconstruction, which is for Dobson the epitome of humanist thinking today (for him, Digital Humanities represents a kind of uncritical return of traditional structuralism). Contrary to the belief that computational techniques can deliver objective truth, Dobson argues that the task of humanists using digital techniques –and once again the use of these techniques is not all considered anti-humanist in itself– is to endlessly question the cultural dimension and thus the subjective and historical determination of their allegedly purely technical and mechanical operations and tools.

In other words: the purpose of Critical Digital Humanities is not to discuss the validity or the interest of the actual results produced by certain computational techniques. It is to question the illusion of objectivity that many associate with this way of working. Dobson therefore rejects for instance the distinction between “data” (supposedly objective and objectifiable) and “interpretation” (inevitably subjective and context bound) that many Digital Humanists accept without further debate. According to him, the fact that one accepts the subjectivity of “interpretation” seems to imply that one also accepts the objectivity of both data and datafication (the transformation of raw materials in units that become open to machine-reading). What interests Dobson in his book is not the discussion whether the interpretation of a digitally, statistically, or logarithmically obtained output is more objective and more reliable than non-digital ways of reading and interpreting (this is more at the center of Da’s analysis, which demonstrates the technical flaws of digitally produced results), but a thorough critique of the pseudo-objectivity of all computer-aided instruments and methods.

In order to make this fundamental point, Dobson always starts by discussing, in a very clear and non-specialized language, the workflow of the Digital Humanities analyses and methods under scrutiny. This presentation allows him to highlight the specific points and moments that are black-boxed, before inviting his reader to open all of these boxes as much as possible––less in order to solve the problem that might occur when the procedure is insufficiently objectified than in order to stress that the problem of “desubjectifying” is insoluble in humanist disciplines. His aim is not to make Digital Humanities more robust, that is less dependent on non-objective choices and decisions, but to contest the very possibility of a completely objective approach that does no longer falls prey to suspicion and critique.

In practice, Dobson illustrates his general thesis by close-reading the general assumptions and practical implementation of three types of Digital Humanities techniques, which he roughly links with the tripartite model of Raymond Williams, one of the founding fathers of cultural studies who elaborated a simple but far-reaching distinction between three types of cultural forms in permanent interaction: residual forms (remains of the past), dominant forms (those that are hegemonic at a certain moment in time) and emergent forms (which may become dominant in a later period). Given the book’s focus on the method of topic modelling, that is the lexical and conceptual categorization of verbal data as treated by machine-reading, Dobson analyzes the gradual shift from initial forms of disclosing “collocated terms” (that is words that tend to appear together) to more sophisticated types of “k-NN” or k-nearest neighbor algorithms (a probability calculation technique to determine the context to which a certain term will be attracted and vice versa) to cutting-edge “neural networks” methods (where the computational tools have been programmed in such a way that they can optimize their own results and procedures in the course of their operation; the specific example chosen is “word2vec”). Dobson certainly acknowledges that each new step offers more precise and interesting information, yet he also stresses the lasting impact of unexamined restrictions and constraints that diminish the claim of complete objectivity.

These constraints have to do with subjectivity and context, such as for instance the underrepresentation of non-mainstream voices in historical documents or, more surprisingly, the imaginary representation of “natural” relationships as articulated by the Zeitgeist, certain ideas of spatial gatherings as seen in suburban and democratic environments where people tend to live with other people they consider similar to them and where conflicts are solved by the rules of democratic majority (if a person X lives in a neighborhood Y where most inhabitants share the feature Y, he or she will statistically be linked with the statistically dominating nearest neighbor, that is Y, not X). But often the constraints are historical in a more radical sense, since they continue to bear the influence of previous ways of thinking (the main example here is the lasting impact of the Bayes’s theorem, which dates from pre-Enlightenment 18th Century but still structures our modern ideas on probabilistic calculation).

Finally, Dobson’s search for a more humanist take on Digital Humanities also has a strong political dimension. The author frequently underscores the role played by certain ideologies (the industrial-military complex is an example of it, but race and gender are equally put on the table) and frequently takes sides with the Foucauldian reading of history, that is defending a critical reading of historical contexts and settings that makes room for alternative ways of thinking (in this case in strong opposition to the computational streamlining and manipulation as made possible by the big data logarithms that tend to program our individual and collective lives).