Style, statistics, and new models of authorship<
University of Newcastle
Hugh Craig. "Style, statistics, and new models of authorship". Early Modern Literary Studies 15.1 (2009-10). <URL: http://purl.oclc.org/emls/15-1/craistyl.htm>.
Whatever the other problems of method and evidence, this astonishing moment of deliberate omission seriously undermines Hoy’s project, and may alert us to the theoretical issues inherent in using “linguistic preferences” and “language practices” in the pursuit of essential and stable identities. These terms, indeed, may expose a problem now more fully legible through the lens of sexuality-theory: is Fletcher’s style chosen or innate? An act or an essence? Are his chosen practices preferred or performative? (17)The implication here is that since style in writing, like gender, is a performance, not an essence, authors are free to vary it at choice and any stylistics that depends on the stability of authorial style is hopelessly compromised. Masten then lists other reasons to doubt that attribution in Early Modern drama based on stable linguistic habits could work. “[C]opyists, actors, [and] compositors” intervene between an author and the texts we have. Collaborators can be expected to influence each others’ styles. Playwrights deliberately write character parts in different styles, in this way “refract[ing] the supposed singularity of the individual in language” (17). What Masten calls “the presumed universality of individuated style” (17) depends in any case, he says, on “a historically inappropriate idea of the author” (18), based in its turn on notions like “intellectual property, copyright, [and] individuated handwriting” which developed after the period in which Beaumont and Fletcher were working (17).
[t]he organisation of memories . . . reflects the person’s own past experience and thought rather than a shared resource of cultural knowledge. While people may remember the same things, they seldom store them in similar associational matrices. The associational matrix seems to me to require that a person’s speech or writing will exhibit a unique idiolect. (178)Authors have often reported that literary creation is an unconscious process, as in the classical doctrine of inspiration. In the early nineteenth century William Blake said he wrote Milton “from immediate Dictation . . . without Premeditation & even against my Will” (qtd. in Bennett 61). Cognitive science has rediscovered this idea.
 This paper was given as a keynote address at the 2008 Resourceful Reading conference at the University of Sydney. I am grateful to the organisers, Katherine Bode and Robert Dixon, for inviting me to the conference. I am also grateful to the University of Newcastle (NSW) Writing Cultures Research Group, and to Rosalind Smith, for very helpful comments on the paper in its earlier forms.
 E.g. Burrows. “’I lisp’d’”, “Computers.”
 E.g. Craig, “Is the Author” and Forsyth, Holmes and Tse.
 E. H. C. Oliphant and E. K. Chambers both thought Fletcher had modified his style in his part of Henry VIII (Vickers 347). There are no signs of this “accommodation” in the function-word data I have analysed, as discussed below. In other studies, not reported here, I did find some convergence between Fletcher and Shakespeare in lexical word patterns in the play.
 It is worth mentioning that McMullan’s critique is not directed to computational stylistics as such, but rather at the wider movement of scholars using all kinds of quantitative measures in attribution. He notes in the 2000 Henry VIII introduction that computational methods are not yet widely accepted (“because of incompatible fields of understanding both about the nature of statistical study and about the significance of ‘literary’ concepts such as genre”) and anticipates better results once more extensive electronic corpora of the plays are available (193n).
 These are the twenty-eight plays listed as fitting “a minimal definition of the Shakespeare canon, excluding all works of doubtful or collaborative status” in a table in the Oxford Shakespeare Textual Companion (Wells, Taylor, Jowett and Montgomery 81 and Table 2). From this group I have excluded Measure for Measure, which is now generally regarded as a collaboration with Middleton (Jowett 681-2).
 The words are
are again all as being can dare did do doth each ever hath hence in (preposition) itself only may might more must my now of quite rather still that (conjunction) the there these those to (preposition) too very which (relative) who (relative) with ye yetFor each of these word-variables the probability that the Shakespeare and Fletcher character parts belonged to the same parent population was less than 0.0005, according to the t test. Function words are the best understood variables in computational stylistics, having been used regularly for authorship attribution since Frederick Mosteller and David L.Wallace’s work on the Federalist papers in the 1960s.
 For a full presentation, see Chatfield and Collins (on the method) and Burrows and Craig (on applications to stylistics). Calculations were performed with SPSS 16.0, analysing the correlation matrix and using an unrotated factor solution.
 McMullan follows Masten in noting that both Hoy, and Jonathan Hope in his Shakespeare-Fletcher study, omit The Faithful Shepherdess from their Fletcher control sets because this play is so unlike the rest of the Fletcher canon in style. McMullan regards this as more evidence that attribution studies are inevitably compromised by the variation within authorial canons (Politics 451).
 He adds: “It is inherent, however, not merely in statistical principle but in human behaviour at large, that such evidence cannot be absolute. The consistencies we observe are trends, not universals” (28-9).