Troy: The Musical? How Memnon Could Have Sung

08 Aug, 2026

Homer was sung, and the technique for reconstructing that song fits Memnon — the one Trojan story Homer never tells — using the Iliad's own formulae for Hector's death and Thetis's grief; here is an attempt to experiment with it.

This essay is part of an ongoing Homer project asking what ancient evidence, and modern tools, can and cannot tell us about the epics. Earlier installments include: Animating Homer: Teaching AI the Visual Language of Ancient Greece

Detail of a Mycenaean krater showing a lyre player, Archaeological Museum of Nafplio

Krater with a lyre player, detail. Late Helladic III, 1350–1250 BCE, from chamber tomb IV at Evangelistria, Nafplion, in the Argolid. Painted some five centuries before the epics reached written form, the seated figure is among the earliest Greek depictions of the instrument the poems place in a singer's hands, and it comes from the Mycenaean world the epics remember rather than the Archaic one that wrote them down. The shape is a chelys, the tortoiseshell-bodied lyre, rather than the phorminx Homer names, which is a useful reminder that the instrument behind Homeric performance is reconstructed from scattered evidence rather than settled by any single image. Collection of the Archaeological Museum of Nafplio, inventory number 202181. Photograph taken 25 September 2020 by Zde, own work, licensed under CC BY-SA 4.0 and available at full resolution (3,333 × 2,218 pixels) via Wikimedia Commons.

Homer is read in silence, on a page, usually in translation. For most of the tradition's working life nobody encountered it that way. Heroic narrative reached its audiences as performance: a singer, a stringed instrument, a hall, an audience that already knew the plot and had come for the telling. The written epics are a late artefact of a system that ran on voice, and no musical notation survives from any of it.

Which makes the following recording remarkable. Stefan Hagel of the Austrian Academy of Sciences, singing the opening of the Iliad on the Pnyx in Athens in March 2017, accompanying himself on a lyre.

Stefan Hagel performs the proem of the Iliad on the Pnyx, Athens, 4 March 2017. Part of the reconstruction technique developed with Georg Danek of the University of Vienna.

Most modern listeners find it austere on first hearing, and repetitive, and shorter on drama than the recorded readings through which Homer usually reaches us. The austerity is the point. What is being reconstructed is not a composition but a delivery system, and understanding how that system works turns out to open a door onto the one Trojan story Homer knew and declined to tell.

What You Are Actually Hearing

Georg Danek and Stefan Hagel spent years developing a technique for singing Homeric hexameter, and the careful thing about their claim is where they place it. Their recordings are not to be understood as the exact reconstruction of a given melody. They reconstruct the technique by which a singer accommodated melodic principles to the demands of an individual verse, which is a different and more defensible object.

The mechanics rest on a property of the language. Ancient Greek was pitch-accented, so every word carried its own melodic shape before any singer touched it: an acute raising the accented syllable, a circumflex rising and falling within one long syllable, everything else sitting at the level of ordinary speech. Set that language inside dactylic hexameter and much of the melody is already determined. A singer works from a broad contour spanning the verse, inflects at the caesura, adjusts locally so each accent lands as a peak, and descends at the line's end. Because the contour is regenerated line by line, and performance by performance, the melody was improvised alongside the words rather than recalled. No two tellings needed to match.

None of which would carry much weight if the practice itself had to be inferred from the outside, and it does not. The most useful evidence for Homeric performance sits inside Homer. Book 8 of the Odyssey stages a singer, Demodocus, performing Trojan War material before an audience that includes Odysseus, and the poem describes the occasion with the attentiveness of a professional describing his own trade: the instrument brought out and hung on its peg, the singer led in and seated, the audience quieting, the hero weeping behind his cloak at a story about himself. Phemius performs under duress in Ithaca. Achilles sings to a lyre in his tent while the embassy waits. The epics are, among other things, documentation of the practice that produced them, and what they document is not recitation but composition in performance.

The instrument is the remaining variable, and it is genuinely open. Danek and Hagel's published work centres on the phorminx, the four-stringed lyre depicted in the poems and on archaic pottery, which restricts the available pitches to a tetrachord and produces the narrow, chant-like quality of the recordings. In other performances Hagel uses a seven-stringed lyre in what may have been an archaic Aeolian tuning, which widens the melodic range considerably. The choice is a modelling decision rather than a discovery, and it changes the sound more than any other single assumption.

A longer stretch: Iliad 14.341–388, the deception of Zeus. Sustained narrative rather than a proem, which is where the improvisatory character of the technique becomes audible.

The Song Homer Doesn't Sing

Every episode Hagel performs comes from a poem we still have. The interesting application lies elsewhere, among the stories the surviving epics assume their audiences already know.

Memnon is the strongest of them. Son of Eos and the mortal Tithonus, he arrives at Troy as an ally after Hector's death, kills Antilochus the son of Nestor, and is killed in turn by Achilles, whereupon his mother mourns him and secures immortality for him from Zeus. Homer never narrates any of it, yet twice lets it surface. In Book 4 of the Odyssey, at Menelaus's table, Nestor's surviving son cannot keep from weeping as he remembers the brother whom the shining son of bright Dawn had killed. In Book 11, in the underworld, Memnon appears as nothing more than a standard of beauty against which another hero is measured. Hector's dying prophecy in Iliad 22 sets Achilles' own death at the Scaean Gates by Paris and Apollo, a scene the Iliad likewise declines to show.

The architecture is what makes the case compelling rather than merely tantalising. Antilochus dies, Achilles takes the field to avenge him, Memnon dies, Achilles moves toward his own death; set that beside Patroclus, Achilles, Hector, and the same ending, and the two sequences are the same sequence. Scholars have argued the relationship in every available direction for a century, whether Homer reshaped an older Memnon tradition, whether the lost Aethiopis reshaped Homer, or whether both drew on story patterns available to any singer who wanted them. What matters for a reconstruction is the shared conclusion underneath the disagreement: the Memnon material is old enough, and traditional enough, to have been sung.

Nobody should pretend a poem called The Song of Memnon is attested, because no such title survives and no manuscript records one. The defensible claim is that the wider Trojan tradition circulated as performed heroic lays and flexible story patterns long before cyclic poems fixed particular versions in writing, and that Memnon was among the material in that repertoire.

What the Surviving Music Is Fit For

The rules that govern the reconstruction have to come from somewhere, and the corpus they come from is thin, late, and generically remote from epic. A papyrus written around 200 BCE preserves notation for a few lines of the first chorus of Euripides' Orestes, a play first performed in 408 BCE, and scholars dispute whether the melody is Euripides' own or a later setting. Two hymns to Apollo, inscribed on the wall of the Athenian Treasury at Delphi in the late second century BCE and recovered in 1893, preserve extended vocal and instrumental notation. The Seikilos inscription preserves a short later melody essentially complete. Several compositions attributed to Mesomedes add examples from the Roman period.

None of that is Homeric music, and the gap is not marginal. The Delphic Hymns postdate the composition of the epics by something like six centuries and belong to a different genre, a different occasion, and a probably different instrumental practice. Reading a Homeric melody out of them would be an ordinary fitness-for-purpose error, the kind that arises whenever a dataset is used to answer a question it was never collected to answer. A figure being real does not make it the right figure, and a melody being genuinely ancient does not make it evidence for a tradition six hundred years older.

What the corpus is fit for is narrower and considerably more valuable. The notation shows how Greek composers set Greek words: how melody responds to acute and circumflex accents, how note duration tracks syllable quantity, how closely musical phrasing follows linguistic phrasing, and what melodic behaviour recurs at phrase endings. Where those relationships hold across the Orestes fragment, both Delphic Hymns, Seikilos and Mesomedes, they start to look like properties of the language and its musical treatment rather than the signature of one composer or one century. Persistent relationships of that kind are legitimate constraints on a reconstruction of earlier practice. A single hymn's tune is not, and the distinction between the two is what separates this exercise from a soundtrack.

A Worked Example: The Death of Memnon

Here is what the technique produces when pointed at that repertoire. The passage is the sixth and seventh movements of the sequence, Memnon falling and Eos mourning, and it can be built almost entirely from lines the Iliad already contains.

Pitches are numbered one to four from the bottom of a four-string tetrachord. Two is the base recitation note. Accented syllables rise to three, four is held for the peak of a verse, and the line cadences to one. Superscripts sit above the syllables in imitation of the ancient practice of writing notation above the vowels.

Memnon's death, in the verse that closes Hector at Iliad 22.361:

ὣς³ ἄ³-ρα² μιν² εἰ²-πόν³-τα² τέ³-λος² θα²-νά³-τοι²-ο² κά³-λυ²-ψε¹

Thus, as he spoke, the end of death enfolded him.

Quantity runs – u u | – – | – u u | – u u | – u u | – u. The circumflex on ὣς falls from three to the base inside its own syllable. Each of the four acutes takes three, everything between them stays at two, and the final syllable resolves to one. Not a single pitch in the line is a modern aesthetic decision; the melody is a function of the accents and the cadence, which is precisely Danek and Hagel's point about where the singer's freedom actually lay. The following verse, ψυχὴ δ' ἐκ ῥεθέων πταμένη Ἄϊδόσδε βεβήκει, sends the soul flying from the limbs down to the house of Hades on the same cadence, which is how a run of hexameters coheres without any fixed tune.

Then the mother. Eos can be given Thetis's lament for Achilles at Iliad 18.54, word for word:

ὤ³ μοι² ἐ² γὼ² δει² λή³ ὤ³ μοι² δυ² σα² ρι² στο² τό³ κει² α¹

Woe is me, wretched, woe, unhappy mother of the best of sons.

The line is slow by construction rather than by instruction. Epic correption shortens μοι before ἐγώ, which is why the opening scans as a dactyl; two spondees follow in the second and third feet; and the extraordinary compound δυσαριστοτόκεια, a word Homer uses once in all of epic, carries a single accent across five syllables. Anyone reconstructing this movement would want slower delivery and longer tones, and the metre supplies both without being asked.

Which is enough to render, so here is the passage: the tetrachord sounded on the strings, then the two verses of Memnon's death, then the lament.

A synthesised realisation of Iliad 22.361–362 and 18.54. Pitch derives from word accent, duration from syllable quantity, and the four available pitches stand in for the phorminx. The voice is formant synthesis rather than a singer, so it carries the melodic system and the reconstructed vowel qualities while telling you nothing about how a human being would have sounded performing them.

Set it beside Hagel's recording above and the difference is instructive. His has a person in it, with all the phrasing decisions a person makes, and the gap between the two is exactly the gap between a system and a performance.

What Falls Out of the Choice

The result is more interesting than a demonstration of method. Memnon's death can be sung in the Iliad's own death formulae, and his mother's grief in Thetis's own words, with no adjustment beyond a name. Both mothers are goddesses. Both sons are half-divine, doomed, and mourned before they fall. Both deaths are the hinge on which Achilles' approaching end turns. The formulaic apparatus for the entire scene already exists in the surviving text, fully formed, and it fits Memnon without strain.

The fit is not a convenience to be quietly enjoyed. It is the scholarly argument arriving as a practical result. A tradition that composed in formulae had the components for this scene in stock, which is exactly what one would expect whether Homer worked from an older Memnon song, or a later poet worked from Homer, or both singers reached into the same inherited store. A reconstruction cannot adjudicate between those accounts. What it can show, concretely, is how little new material the scene requires, and that is itself evidence about how such songs were made.

Which raises the question of how a reader is supposed to tell the inherited material from the modern joinery, and the answer has to be structural rather than a disclaimer at the foot of the page. The workable approach is a confidence hierarchy applied line by line and kept visible in the finished artefact.

At the top sit the narrative facts securely preserved in the ancient tradition, the six events of the Memnon sequence and nothing beyond them. Below that sits attested Homeric language, whether complete verses like the two quoted above or individual cola recombined as the tradition itself recombined them: formulae for dawn, arrival, arming, spear combat, death, grief, divine mothers and fate, all drawn from the surviving corpus. At the bottom sits reconstructed connective material, composed only where the attested elements will not join, metrically valid, and marked unambiguously as modern work. In the passage above, both notated verses and the soul-departing line belong to the second tier; the transition in which Eos sees her son fall does not yet exist, and its eventual replacement should be legible as modern to anyone who wants to check.

Eighty to a hundred and twenty lines built that way are not a lost poem recovered. They are a demonstration of what the tradition's own materials permit, with the provenance of every line available for audit, which is the only form in which such a reconstruction is worth anything at all.

Many Performances, Not One

The final property of the model is the one most likely to be misunderstood, so it is worth demonstrating rather than asserting. A single verse admits several realisations that all obey the rules. Here is Iliad 22.361 three ways.

ὣς³ ἄ³-ρα² μιν² εἰ²-πόν³-τα² τέ³-λος² θα²-νά³-τοι²-ο² κά³-λυ²-ψε¹
ὣς³ ἄ⁴-ρα³ μιν² εἰ²-πόν³-τα² τέ³-λος² θα²-νά³-τοι²-ο² κά³-λυ²-ψε¹
ὣς² ἄ³-ρα² μιν² εἰ¹-πόν³-τα² τέ²-λος¹ θα¹-νά³-τοι²-ο¹ κά³-λυ²-ψε¹

The first is the plain accentual realisation. The second lifts the verse opening to four and relaxes back, a natural choice for a line beginning a narrative unit. The third narrows the register after the caesura, dropping the base to one so the accents stand out more sharply, which suits a verse turning into a lament. All three keep every accent as a local peak, hold every unaccented syllable at or below the base, stay inside the tetrachord, and cadence to one.

The same verse three times: identical text, identical rhythm, three legitimate performances of it.

Which is where computational tools become genuinely useful, and where they should be kept. A model can trawl the Homeric corpus for formulaic parallels far faster than a reader, test candidate lines for metrical validity, analyse pitch accent across long stretches, and generate rule-compliant melodic realisations by the hundred. Filter a hundred outputs down to the ten that survive scrutiny, record three from opposite ends of the range, and the deliverable is a plausible space of performances with its boundaries stated. A system that produced one definitive melody would be making a claim about the tradition that the tradition contradicts, since the singers themselves never produced the same performance twice.

There is a resemblance here worth noticing without overselling. The ancient singer worked from an inherited story, a stock of formulae, a metrical constraint and a set of musical conventions, and produced one particular performance. A reconstruction works from a surviving outline, the same corpus, the same metre, the surviving musical evidence and a set of computational rules, and produces many. Nobody should claim a model composes as an aoidos composed. The resemblance does explain why variation, not resolution, is the honest output.

Greek audiences met these heroes through voice, rhythm, melody and a plucked instrument, and that encounter is the part of the tradition for which nothing at all survives. The missing books of the Epic Cycle are the loss everyone counts. The missing register may be the larger one.

What survives is enough to constrain an experiment: Homer preserves traces of stories his poems never narrate, later summaries preserve their plots, and a small corpus of later notated Greek music preserves enough of the system to rule some reconstructions out. The aim is not to make Memnon sing as he did three thousand years ago, since that target is unreachable and any claim to have hit it would be false. The aim is to establish, as rigorously as the evidence allows and with every assumption on the record, how Memnon could have sung.

My blog