Homer's two epics run to nearly 28,000 lines, and were composed without writing, at speaking speed, in performance, by a poet who did not have them by heart. The way that is possible is the subject of this page. Milman Parry's answer, in 1928, was that the verse is not remembered but woven, live, from ready-made, exactly-measured word-groups he called formulae. Here you can weave a real line of the Iliad the way a bard did, and watch it scan.
Set a clay tablet beside a guitar. The tablet is how we imagine a poem is made: a text, fixed, copied, learned. The guitar is closer to the truth about Homer. A folk guitarist does not memorise a three-hour set note for note; they know chords, runs, turnarounds, the shapes, and assemble the song as it goes. Milman Parry's discovery is that Homeric Greek gave its singers a stock of shapes so precisely engineered that a poet could build a flawless six-beat line, on the fly, out of parts that were guaranteed to fit.
Two things have to be true at once for this to work, and both can be shown rather than asserted. First, the line has a rigid metrical shape (dactylic hexameter) with only a little give. Second, the tradition supplied word-groups pre-cut to fill the gaps that shape leaves. Get both, and composition stops being memory and becomes fitting: like a weaver who has dyed the threads to length before the shuttle ever moves. Start with the loom's frame.
A hexameter is six feet. The first four are each a choice between a –⏑⏑ (a dactyl: one long, two shorts) and a –– (a spondee: two longs) – same duration, since a long counts as two shorts. The fifth foot is almost always a dactyl; the sixth is always two syllables, the last of them free. Click the first four to feel the give:
The other fact is quieter and does all the work. Almost every hexameter has a word-break, a caesura, near the middle, in the third foot, so the ear hears the line as two halves (M.L. West: "we may regard the hexameter as consisting of two cola"). In Iliad Book 1's 611 lines the break falls after the first short of the third foot (the "feminine" caesura) in 356 lines and at the strong "masculine" break in 247. And, West notes, it is the feminine break that "is more often presupposed by formulaic phrases." That is the seam the singer weaves across: build the first half up to the break, then drop in a subject-phrase cut to exactly fill the rest.
Caesura counts for Book 1: Monro, Homeric hexameter, via Dickinson College Commentaries. The two-cola / caesura statements: M.L. West, Introduction to Greek Metre (1987), 19–22.
Here are two racks of real threads from Book 1 of the Iliad. The gold rack holds opening half-lines: speech-introductions that run from the line-start to the caesura ("and answering him he addressed…"). The teal rack holds line-final namings: a hero or god plus a fixed epithet, cut to finish the verse. Pick one from each and the loom joins them, scans the result live, and tells you whether that exact line stands in Book 1, or is simply a well-formed hexameter the system could have produced.
The verdict is honest: a line is only marked attested if those exact words stand at that line of Book 1. Every other valid combination is a hexameter the technology licenses (the kind of line Homer could have sung), never a line we claim he did.
Notice what the loom is teaching as you play. The opening half-line leaves a gap of a certain size; only a naming cut to that size will complete a legal hexameter. Swap πόδας ὠκὺς Ἀχιλλεύς (swift-footed Achilles) for νεφεληγερέτα Ζεύς (cloud-gathering Zeus) after the same opening and the line still scans, because both namings were built to the same slot. Reach for Agamemnon's long form after that same opening and it overflows the line: the singer would have reached for his short Agamemnon instead. That is the whole trick. The poet is not recalling lines; the poet is fitting threads to a gap.
Parry's definition of the unit is the most-quoted sentence in Homeric scholarship, and the loom is a working model of it:
a group of words which is regularly employed under the same metrical conditions to express a given essential idea.Milman Parry, defining the formula · "Studies in the Epic Technique of Oral Verse-Making I" (1930)
"Under the same metrical conditions" is the gold and teal racks: each thread has a fixed metrical shape and a fixed place it belongs. "A given essential idea" is why the singer can treat them as interchangeable spare parts. The essential idea of every gold thread here is just he answered; everything else is the metre's business, not the meaning's. Parry's own example of a formula was one of these very half-lines, τὸν δʼ αὖτε προσέειπε, whose essential idea, he wrote, is merely "he said to him."
Two measured properties turn a pile of threads into a system. Parry named them extension (the system reaches across all the characters and all the sizes of gap a line can leave) and thrift, or economy:
The thrift of a system lies in the degree in which it is free of phrases which, having the same metrical value and expressing the same idea, could replace one another.Milman Parry, defining thrift · Collected Papers, 276
Read plainly: for a given character, in a given slot, the tradition keeps one thread, not a drawer of synonyms. A drawer of synonyms would be lovely for a writing poet reaching for variety; it is dead weight for a singer who needs the right-sized part now. Parry's headline count: "in a list of 37 characters who have formulas of this type… there are only three names which have a second formula which could replace the first." You do not have to take his word for it. The economy is visible in the text, so let us count it.
Below is every time one of seven principal figures is named at the end of a line in Book 1 of the Iliad – recounted from the Perseus text by the same scanner that runs the loom. For each figure the recurring name-epithet formulae are grouped, with their metrical shape (– long, ⏑ short) and how often each appears; ordinary namings that are not a stock formula are pooled at the foot as "one-off."
The shape of the result is Parry's, made of the text's own bricks. Each figure carries a small kit of formulae of different metrical lengths: that is extension, the singer's ability to name Achilles whether he has three syllables to fill or seven. And within any one length there is essentially one formula, used again and again, and that is economy. Achilles is πόδας ὠκὺς Ἀχιλλεύς to fill the long slot and δῖος Ἀχιλλεύς to fill the short one, and almost nothing else. Then there is the honest remainder: the "one-off" lines, where the poet names the hero not with a formula but with ordinary syntax. The system is a strong current, not an iron law, which is exactly the next thing to be honest about.
This is a genuinely contested corner of scholarship, and the loom would be a lie if it hid that. Four edges, each real:
Parry proved the diction is formulaic and traditional. That the poems were therefore composed in performance by a non-writing singer is an inference he added later, and he knew it was not a proof: "If one wishes to think that Homer composed his poems orally, and then sat down and wrote them out, there is little that can be said in disproof, and little that needs to be said." How our fixed texts actually came to be (pure performance, dictation to a scribe, a literate master using traditional craft, gradual crystallisation over centuries) is still open. Treat oral composition here as a strong, well-supported hypothesis, not a settled fact.
Even the count above shows it. Agamemnon carries two long formulae of almost equal metrical value, ἄναξ ἀνδρῶν Ἀγαμέμνων and εὐρὺ κρείων Ἀγαμέμνων, where strict thrift wants one. Parry flagged such doublets himself; David Shive (Naming Achilles, 1987) argued there is more choice and less economy than Parry claimed, J.B. Hainsworth (The Flexibility of the Homeric Formula, 1968) showed formulae bend and move, and Rainer Friedrich (2007) counted the breaches and argued many are deliberate art. Economy is real and strong; it is not exceptionless.
The hard reading says the epithet is chosen for metre, not sense: Achilles is "swift-footed" even while sitting still in his tent, because that is the thread that fits, and the audience did not mind. But Anne Amory Parry (Blameless Aegisthus, 1973), Norman Austin, and Paolo Vivante argued the epithets are not semantically dead: they can be contextually alive. The loom shows the mechanism; whether the mechanism empties the words of meaning is not something the mechanism can settle.
Parry and Albert Lord went to Yugoslavia in the 1930s and recorded illiterate guslari composing epic in performance. One, Avdo Međedović, dictated a song of 12,311 lines, Odyssey-scale, from a man who could not read. That proves a poem of Homeric length can be made orally. It does not prove it about Homer specifically, and much of the collected material was itself dictated. Corroboration of possibility, not identity.
Sources for this section: Parry, The Making of Homeric Verse (1971); Lord, The Singer of Tales (1960); Finkelberg, "Oral Theory and the Limits of Formulaic Diction," Oral Tradition 19 (2004); Shive (1987); Hainsworth (1968); Friedrich (2007); A. A. Parry (1973). This line count is the two poems' combined length (Allen edition, 27,793 lines).
Not a claim about the past you have to trust; a claim about this text you can re-run. research/loom-of-the-singer/verify.mjs asserts, from the raw Perseus XML:
perseus-grc2) and matched to that text; the file's SHA-256 is pinned so a future reader sees the same bytes.What is not claimed: that Homer was one person, or illiterate, or that these particular Book-1 formulae exhaust the system. What is shown: that a rigid metre plus a stock of pre-measured, thrifty word-groups is enough to weave a correct hexameter without recalling it, and that the stock is really there, in the text, countable. The tablet was never how the song was made. The loom was.