The Clock Only Euripides Kept

Euripides' plays can be put in order by counting one habit in his verse. Here is the same count, the same program, run on the fourteen surviving plays of Aeschylus and Sophocles. It dates none of them the way it dates Euripides, which scholars have said since the nineteenth century; here you can watch why. Aeschylus's habit moves the other way, Sophocles's hardly moves at all, and the one Aeschylus play that was once dated a generation too early sits with his plays of the 470s and 460s, near where a 1969 argument put it, though the count cannot say on which side of Seven.

The spoken lines of Greek tragedy are iambic trimeters: twelve positions, short and long in turn, and a poet may split a long position into two short syllables. That split is a resolution. Euripides used more of them as his career went on, from about seven in a hundred lines in the 430s BC to more than forty in a hundred at the end. The Clock in the Verse fits that rise to the eight of his plays with a year on record, and tests it by hiding each date in turn: it finds the hidden year to within about five years.

A reader of that page is entitled to ask a plain question. Is the rise a fact about Euripides, or a fact about the Athenian stage in the late fifth century, which every tragedian followed? Two other tragedians survive, with seven plays each, and several of them have dates on record too. So here is the same scanner, unchanged, run over them.

Fetching the text of thirty-three plays…

Look at 409 and 408 BC. Philoctetes, by Sophocles, won first prize in 409. Orestes, by Euripides, was staged the next year. Both were written by men past seventy for the tragic competitions at Athens, and Euripides' play has about four times as many resolutions per line. With names left out, ten of the fourteen plays of Aeschylus and Sophocles sit at or below the rate of Euripides' three earliest dated plays, whatever year they were staged, and none comes near his Trojan Women of 415 or anything he staged after it.

Borrow Euripides' clock

The simplest test is to use the clock exactly as it stands. Fit it on Euripides' eight dated plays, as the other page does, and ask it for the other fourteen:

It puts every one of them in the four or five years around 430 BC. Persians, staged in 472, comes out 44 years late; Oedipus at Colonus, staged in 401, 30 years early. The clock is not a clock of the period. It is a description of one man's habit.

None of this is new, and that is the check

That the clock is Euripides' alone is not a discovery. Scholars counted these resolutions by hand in the nineteenth century and said so. C. F. W. Müller's Latin monograph of 1866 gives totals for all three poets (a note to the Aeschylus total says names are not counted; for the other two he does not say). R. C. Jebb, editing Sophocles' Electra in 1894, printed a count by G. Wolff for each of the seven plays and drew the conclusion this page reaches:

The extraordinarily high proportion in the Philoctetes (409 B.C.) must be considered as indicative of the poet's latest period, and showing the influence of Euripides. But the danger of inference from a comparison of lower ratios is evident. The ratio in the Oedipus Coloneus is lower than in the earlier Tyrannus, and only the same as in the Ajax, which is the oldest play after the Antigone.R. C. Jebb, Sophocles: The Electra (1894), Introduction, §20, n. 4

N. Manousakis put the modern view in one sentence in 2020: The same method cannot apply to Aeschylus and Sophocles. So what this page can offer is not a new verdict but an old one you can operate, and a check on the counter: if an unchanged program, run on different editions, gives the numbers that were got by hand between 1866 and 1896, the rest of the page can be trusted as far as the counting goes.

Müller's totals and this page's agree to within a point in all three poets. For Sophocles the two old per-play counts and this one put the seven plays in nearly the same order, and all three give the same picture: Philoctetes far above, Electra and Antigone at the bottom, Oedipus at Colonus in among the middle plays. Haigh has a reason for that last one: The Oedipus Coloneus, though later in date than the Philoctetes, is much smoother in its versification. But this fact is probably due to the difference in tone between the rapid dialogue of the Philoctetes and the serene tranquillity of the Oedipus Coloneus.

Aeschylus runs the other way, or seems to

Aeschylus has five plays with a year on record, and three of them are one trilogy. Persians was staged in 472 BC, Seven against Thebes in 467, and the Oresteia (Agamemnon, Libation Bearers, Eumenides) in 458, two years before he died. The years come from the production notices copied into the ancient summaries of the plays.

Counting every resolution, Persians and Seven look the same. But both are full of names: the Persian commanders in the messenger's catalogue, the seven champions at the seven gates. A name has to go into the line whatever its syllables are, so it can force a resolution the poet would not otherwise have chosen. Müller in 1866 and Ceadel in 1941 left names out for that reason, and so does this page's main count (a resolution is set aside when either of its two short syllables is in a word the editor capitalised). With names left out, the rate falls from play to play:

Before running any fit I wrote down the test (the pre-registration): the same leave-one-out as the Euripides page, the same five-year bar, names left out. The per-play rates were already on my screen when I wrote it, so it fixes the method, not the answer. Aeschylus passes: the typical miss is under two years, and the correlation between year and rate is strongly negative. His rate falls as the years go on, by roughly one resolution per hundred lines every three years.

That pass is weaker than it sounds, and you should know why. Five plays with three distinct years is very little. The three plays of the Oresteia vouch for each other: leave Agamemnon out, and Libation Bearers and Eumenides are still there saying 458. Leaving out a whole year at a time (a test I added after the pre-registration, so it is only a check) gives a typical miss of about three years. With names counted, it gives about 137 years, because Persians and Seven are then level and a line through two level points can say anything. And the gap between Persians and Seven, names left out, is smaller than twice its own sampling error. What the verse really shows is two levels: three early plays at about nine or ten resolutions per hundred lines, and the Oresteia at about five or six. Manousakis, reading Schein's 1979 tables, describes the same shape: In Aeschylus’ older extant dramas Per., Sev., and Supp. the frequency of resolution ranges from 12% to 8%. In the trilogy it drops to 6%.

The play that was thirty years out

The seventh play is Suppliants: fifty daughters of Danaus, the chorus, flee to Argos to escape marriage to their cousins. For more than a century it was read as the oldest surviving tragedy. The chorus is the protagonist, the actors have comparatively little to do, and that seemed to belong to an age when tragedy had only just grown out of choral song. A. E. Haigh's handbook of 1896 is typical:

The Supplices is undoubtedly the earliest of the seven plays, though its exact date cannot be determined. […] The prominence of the chorus throughout the play, the slightness of the dialogue, and the insignificance of the second actor, are decisive on the point.A. E. Haigh, The Tragic Drama of the Greeks (1896), p. 101

In a footnote Haigh records that K. O. Müller had argued for a date of about 461, from the play's praise of Argos, and says that view cannot be seriously maintained. Estimates of the early date ran back, in Michael Halleran's summary, to perhaps from the fifth century’s first Olympiad, around 500.

Then, in 1952, Edgar Lobel published a scrap of papyrus from Oxyrhynchus, P.Oxy. 2256 fr. 3. It is a production notice. It says Aeschylus won first prize with a group of plays whose legible titles are the Danaids and the satyr play Amymone, the companions of Suppliants; that Sophocles came second; and that a poet called Mesatos came third. Sophocles is usually believed, on a story in Plutarch's Cimon, to have first competed in 468 and won, and Aeschylus won in 467 with the Theban plays, so the production is put between 466 and 459 (the pre-registered test used the slightly wider window 468 to 459). The archon's name at the head of the notice is damaged, and one restoration, Archedemides, gives 463, which most scholars prefer. Not all: some doubt Plutarch's story or its year, and S. Scullion argued in 2002 for 477 to 470. Either way, the judgement from style had been wrong by at least a decade and probably a generation.

Did the dialogue give it away all along? A. F. Garvie thought it pointed the right way. In his 1969 book on the play, after the papyrus, he noted that Aeschylus's resolutions show a steadily increasing restraint, which put Suppliants after Seven and before the Oresteia (that is Manousakis's summary of Garvie's argument; I have not read Garvie's page). Manousakis reports that Scullion showed that this suggestion is false, and that Garvie accepted this in his 2006 edition. One of Scullion's points, as Manousakis gives it, is that eight of the 43 resolutions in Seven fall in thirty trimeters that Eteocles speaks in groups of three between the chorus's lyrics, so that one kind of scene can move a whole play's rate. (Scullion's own case was for the earlier dates of 477 to 470, according to P. B. Cipolla.) I have not been able to read Scullion's own pages, which are behind a paywall, so what follows is not an answer to him. It is the same question put again with the whole count in the open. The test, fixed in advance: date Suppliants from Aeschylus's five dated plays, and measure the uncertainty by resampling. Each resampling draws the spoken lines of every play afresh, with replacement, as if the same poet had written slightly different plays of the same length, refits the line and reads off a year. Ten thousand of them:

The verse puts Suppliants in among Persians and Seven, at about 468, and the middle nine tenths of the resamplings run from 475 to 462. The pre-registration took 490 as its stand-in for the early date, and four of the ten thousand reach it; none reaches 500. That is close to Garvie's observation but not the same. He put the play after Seven; this count's best estimate is just before it, because Suppliants' rate (9.3 per hundred) is a shade above Seven's (8.9), and the two differ by far less than their sampling error (about 1.4 each), so the count cannot order them. What it shares with Garvie is the larger point: if Aeschylus's habit moved steadily in one direction, the play belongs with Persians and Seven and is not twenty years older than Persians. Counting names as well, the same fit gives 465, a check added after the pre-registration. Notice what the interval does not do, though. It runs from 475 to 462, so it takes in both the usual 463 and most of Scullion's 477 to 470. The count cannot choose between the two positions scholars actually hold now.

That if carries the whole weight, and it has to be said as loudly as the result. The line is fitted to plays from 472 to 458. Nothing in the surviving work says what Aeschylus's verse was like in 490, because nothing survives from 490. A poet whose resolutions rose until about 475 and fell after it would produce exactly these numbers with Suppliants in 490. The data here also cannot show a steady trend: Persians and Seven differ by less than their sampling error, so the only difference the count really sees is between the three early plays and the Oresteia. That is the ground on which the argument is weak, and it is visible in the table above without anyone's authority. What the count does show is narrower: in its spoken verse, Suppliants is a close neighbour of Persians and Seven, not of the Oresteia. The resolution rate is no evidence for an early date, but it is not good evidence for a date at all.

Prometheus Bound has no date on record, and whether Aeschylus wrote it at all is disputed. Its rate, names left out, is level with the Oresteia's, and the same line would put it about 457. That settles nothing about authorship: a late Aeschylus and a later writer imitating him would both produce a low count. It only says that if the play is his, the verse does not place it early.

Sophocles barely moves

Only two of Sophocles's seven plays have a year on record. Philoctetes was staged in 409 BC, and Oedipus at Colonus in 401, after his death, by his grandson (the archon year 402/1, whose spring festival fell in 401). Antigone is tied by an ancient story to his election as a general in 441/440, which is not a production record. No line can be tested on two points, so the test fixed in advance was simpler. If Sophocles kept a clock like Euripides', his last play should have the highest rate.

It does not. Oedipus at Colonus sits in the middle, level with Ajax, Women of Trachis and Oedipus the King. What does stand out is Philoctetes. Its rate is well above every other play of his, and a gap that large between one play and the rest would turn up by chance well under once in a thousand shuffles. So Sophocles's count has one fact in it and not a trend: in 409 he wrote one play with an unusually free trimeter, and he did not keep writing that way. Read as a clock, it would put Electra and Antigone first and Philoctetes last, and the date on record for Oedipus at Colonus says that last place is wrong.

Scan a line

All of this rests on a program deciding how many syllables a line has. Here it is, on any spoken line of the fourteen plays. Resolved positions are marked; a resolution inside a capitalised name is marked name and is left out of the main count.

The scanner is the one on the Euripides page, byte for byte (this page imports it from there). It marks vowels long or short by their spelling and by the consonants after them, leaves α, ι and υ undecided, fits the result to the trimeter's pattern, and counts syllables minus twelve. It keeps a line as spoken only if at least three of the four lines around it also scan, and it leaves out lines the editor bracketed as later insertions.