Computer-aided data analysis has become a fixture of 21st-century life: our decision making—our lives—increasingly rely on computer algorithms processing reams of data. So, too, have some of the most basic questions in Classics: what is the relative chronology of Plato’s dialogues? Did Apuleius write the text preserved in a recently recovered manuscript De Platone? And did Aeschylus write the Prometheus Bound?
Manousakis, with the help of a variety of computer-aided calculations, shows that Aeschylean authorship of Prometheus Bound is very highly unlikely. Mark Griffith had already made the case against Aeschylean authorship, and after the publication of Griffith’s book in 1977 (The Authenticity of the Prometheus Bound), the burden of proof was now with those—few in number to be sure—who would argue that Aeschylus wrote Pr. Manousakis book’s main conclusion, then, that “the author of the Prometheus Bound could not have been Aeschylus”, may not surprise anyone; but those who read it will agree that Manousakis has ensured that that surprisingly resilient ghost—that Aeschylus may just possibly have been the play’s author—will no longer haunt scholarship. A subsidiary argument for the dating of the Prometheus Bound to the decade 440-430 BCE convinces, though cannot be conclusive.
So, there are the two main arguments Manousakis pursues, on authorship and dating. But his book is valuable also for observations that have little to do with Aeschylus. Again, employing computational methods, the author provides further support for long-held views (e.g. the Aspis is not by Hesiod; the dating of Euripides’ plays, based on increasing use of resolution in trimeter verse); offers insight into persistent questions, like whether the Iliad or Odyssey are written by one person (Manousakis’ analysis suggests two different authors); reveals suggestive linguistic affinities (between e.g. Euripides’ prosatyric Alcestis and his satyr play Cyclops). Some of the tests Manousakis runs group together plays—not always reliably—we know otherwise to be close together chronologically. What are the data Manousakis submits in order arrive at these results?
They are of two kinds. The most important are linguistic traces, which, like a fingerprint, point to a unique hand. The traces will preferably be items that do not ‘mean’ anything. They can be function words like conjunctions, prepositions, (in)definite articles, and so on; or phonemes/morphemes, typically two or more characters in length. Preferred are phonemes that do not map onto a content-word. Content words (“car”, “road”, “grass”, for instance) could very well be subject to authorial control, their choice influenced by the topic discussed, or by the genre. These form part of the “style” of an author—the area of language more or less under his or her conscious control. Manousakis treats eight stylistic features total, covering the playwright’s lexicon (e.g. lexical richness and redundancy; that is, the lexically diversity or parsimony of the author’s vocabulary); metrical practice (e.g. interlinear hiatus), and syntax (e.g. alternation between ὡς and ὅτι complement clauses); among others.
In a succinct two-page introductory chapter (Ch. 1) Manousakis warns us that the book “represents rather demanding reading”. While his monograph will speak most to a specialist readership—one already familiar with the methods used—Manousakis nevertheless tries hard to make his monograph accessible to neophytes, offering a useful overview of the history of authorship and chronology studies (in Ch. 2), concisely explaining the computational methods when introduced, and concluding the whole with a generous bibliography.
The third chapter (Ch. 3), on stylistic features in the Prometheus Bound, reviews and advances previous scholarly discussion. For instance, in the section on sentence length, the author presents previous researchers’ inconclusive results on sentence length (as a variable distinguishing Pr. from other plays). He then presents his own investigation on sentence length in choral odes, where indeed, the mean sentence length for choral odes in Pr. differs significantly from that in the choral odes of securely attested tragedies of Aeschylus (p. 140). For another example, Manousakis reviews for us the long-noted preference of tragedies for ὡς over ὅτι. Others had observed, on partial evidence, that the author of Pr. comes close to Sophocles in the use of ὅτι clauses. Manousakis reviews the material anew. He finds that in all extant Aeschylus, there is only one ὅτι clause after a verb of knowing, understanding, showing, and saying; but “there is not one extant play by Sophocles with fewer than two ὅτι complement clauses” (123, emphases mine). The Prometheus Bound contains seven such clauses, and these are often wielded in a Sophoclean manner. In Sophocles, ὅτιcan appear as the last word in the trimeter line, leaving the hearer in momentary suspense; the gnomic statement which follows is later on contradicted, or turns out not to indicate the speaker’s actual state of mind (see most notably Ajax’ great monologue in the eponymous play, specifically at 678-682). Similarly, the author of Pr. “often marks…gnomic utterances, introduced by ὅτι or other conjunctions (though never ὡς) by using the micro-suspenseful technique of ‘Sophoclean enjambment’.” All but one of these gnomic statements are, in a Sophoclean manner, challenged in some way in the subsequent few lines or in the course of events (pp. 131-132).
We have only six, possibly seven plays from a playwright whose output reached a total of ninety plays. So, although the arguments based on stylistic evidence suggest a poet other than Aeschylus, the possibility remains that an artist of Aeschylus’ caliber changed his style over time or experimented with it at a given point in his career. Manousakis, then, turns to more secure evidence—the analogue to fingerprints (if not DNA) in criminal investigations—linguistic traces. Of the two categories, function words and character n-grams, the latter, consisting of phonemes in 2 or more characters in length, supply “the widest and most thorough representation of authorial trace possible” (221).
The strength in Manousakis’ research lies, partly, in the multiplicity of methods used. One of these on its own could not make a convincing case, but all five of the methods applied demonstrate that the Prometheus Bound’s ‘fingerprint’ or ‘trace’ is not Aeschylean: from Principal Components Analysis performed on function words to the Support Vector Machines (SVM) algorithm, by which the computer is first ‘trained’ to recognize the trace of each in two distinct corpora (say Aeschylus and Sophocles), with which training the machine can then class ‘new’ examples into one or the other group.
A further plagiarism algorithm suggests that the author of Pr. was “one and one sole”, providing an argument against multiple authorship of the tragedy. To simplify, Manousakis compares the mean frequency of a given character n-gram in a section of text against its mean in the entire text and repeats this process across several sections of the text under examination, and for a number of different n-grams. The sum of the differences for all the n-gram variances results in a kind of difference measure which can tell us where and to what extent the text deviates from a putative authorial norm. The algorithm returns no “suspect fluctuations” in the Pr. which might have suggested different hands.
The accumulated results lead Manousakis to suggest, following West (Studies in Aeschylus, 1990), that Euphorion, one of Aeschylus’ sons, authored the play. Euphorion, according to one testimony, had beat out Sophocles and Euripides for first prize at the City Dionysia in 431 BCE, and restaged some of his father’s plays. But another of Aeschylus’ sons, Euaion, who acted in both his father’s plays and a couple Sophoclean ones, could also have composed the Pr. Ultimately Manousakis soberly admits that “in all likelihood we will never know who its author actually was.” His book is indeed difficult reading but constitutes a valuable entry in the field of Aeschylean studies and Latin and Greek computational/corpus linguistics.