De Gruyter (2020) h/b 282pp £109 (ISBN 9783110687644)

In 1977 there appeared Mark Griffith’s The Authenticity of Prometheus Bound, which gave detailed and convincing arguments against Aeschylean authorship: so powerful, indeed, that Martin West was wholly persuaded. Among leading anglophone scholars, perhaps only Hugh Lloyd-Jones continued to believe that Aeschylus composed the play. The book under notice here powerfully—indeed decisively—backs Griffith’s case. It is not, however, easy reading: for, as M. says, it will be necessary for the reader to become familiar with the terms and methodologies of statistics and computer science (there are several deeply daunting equations, and there is much more; for example, your reviewer had to come to terms with cosine similarity). It will be convenient here to note that a French scholar, M.P. Pattoni (1987) disputed Griffith’s findings: M., however, comments that ‘she barely managed to mitigate the broad influence of Griffith’s sound inferences, as has become clear over the years’. 

The headings of the book are as follows: General Conclusions; Prometheus and the Athetesis Question; Quantitative Style in Prometheus; Applying Automated Authorship Attribution in Greek Tragedy: The case of Prometheus; and Who composed Prometheus? A Possibility. (Then come two appendices, a bibliography, and an index nominum et rerum; there are also 30 figures). The two crucial aspects examined by M. are style and trace: style ‘indicators describe the various nuances of the crafted form of a work of literature, while ‘trace’  indicators provide one with meta-knowledge … ‘trace’  detection is external to the content of the text, while stylistic analysis is internal’. (M. lists eight Greek words or terms denoting literary style generally, though not the word particularly applicable to Aeschylus, ὄγκος, τὸ ὀγκῶδες).

In the case of style, many aspects relating to the question of authenticity of Prometheus are ancient, going back as far as Michael Psellos in the eleventh century; however, by using such modern tools as ‘corpus linguistics’, further progress can be made: thus one can now indicate with ‘quantitative’ confidence that the author of Prometheus, unlike Aeschylus, avoids the repetition of words within a few lines of each other, and that the vocabulary of the disputed play is closer to Sophocles than to Aeschylus; again, the use of ὅτι and ὡς in complement clauses is Sophoclean rather than Aeschylean.

But the ‘meat’ of the book lies in the ‘state of the art techniques’ which enable the reaching of valuable conclusions about the un-Aeschylean nature of Prometheus via automated authorship attribution techniques: i.e. ‘trace’, ‘an unintended and rather “uncontrollable” mark left by the author in a text’. To this end comes the computer-based statistical evidence: M. refers to ‘cutting-edge’ computer methodologies in authorship attribution, allowing access to the ‘sub-atomic structure’ of a text. The ‘traces’ are n-grams of various kinds—words, parts of speech, syntactic—or revolving sequences of phonemes-letter strings (character n-gram): the latter provide the most accurate results. N-grams are of various lengths (bigrams, trigrams, tetragrams etc) according to the needs of each individual study. It is no surprise that the preparation of the texts for the experiments was highly time-consuming: six plays by Aeschylus, and six by Sophocles, Euripides and Aristophanes (all texts being homogenised by M., rather than the texts of TLG: this must have been a case of labor improbus indeed, to remove all features for which a given editor is responsible). 

The textual feature used by M. to determine whether the text of Prometheus was revised or manipulated by someone other than Aeschylus is the normalized frequencies of character n-grams: here your reviewer passes over the language of statistics and the equations, to reach M.’s conclusion that PV is indeed by a single author. Thereafter, M.’s text becomes increasingly technical (e.g. ‘the variables [are] eigenvectors of the covariance matrix of the data’). 

Some interesting results include the easy differentiation of Lycophron’s Alexandra from tragic and comic texts, and the close relationship between Alcestis and Cyclops, implying a generic affinity, the indication that the revised Clouds appeared about ten years after the first performance, and even—via Burrows’ delta procedure—that Iliad and Odyssey were ‘likely crafted by different poets’. In the case of Prometheus, the most sensitive broadly tested authorship attribution model (SVM = Support Vector Machines) has shown that character ngrams ‘trace’  of the author of Prometheus is in the main Sophoclean (which of course in no way implies that Sophocles wrote it! The robustness and sensitivity of the techniques used are fully demonstrated by the fact that at one point in Ecclesiazusai Aristophanes seems to have become Euripides—because it is precisely where Euripides is being parodied by Aristophanes). Indeed all five analyses (Burrows, SVM, CMG, PCA, Cluster) employed by M. have the same outcome: Prometheus is extremely unlikely to be the work of Aeschylus, and it may have been composed between 440 and 430 BC. M. finally, tentatively, suggests that a son of Aeschylus—Euaion or Euphorion—might have been the author: Euaion appeared in at least two plays by Sophocles.

To sum up: Griffith’s work is here amplified and strengthened by the use of sophisticated computer techniques to reach the identical conclusion: Prometheus Vinctus is not the work of Aeschylus. It is not difficult to think of other disputed texts (e.g. Plato’s Seventh Letter, some orations of Lysias, some of Ovid’s Heroides) which would benefit from similar analysis: as for Rhesus, surely the work of Liapis has rendered further work unnecessary. 

In the nature of things, it is likely that this book will be more often referred to than read, which in no way reduces its high degree of scholarly usefulness: the editor deserves our thanks and congratulations. As always with de Gruyter, presentation is exemplary. 

Colin Leach