This article and the one it is based on have been summarized in video format using Google NotebookLM.

During my middle school studies in France, I was introduced to Homer’s Iliad and Odyssey. This year, I decided to revisit these epics and acquired a French edition from the Bibliothèque de la Pléiade (translated in 1955 by Victor and Jean Bérard, and Robert Flacelière). This particular edition is noteworthy for its clear demarcation of dialogue: each speech is prefixed with the speaker’s name, as illustrated by the following excerpt available on the La Pléiade website (see pages 12-13 for example).

The editorial practice of prefacing each speech with the interlocutor’s name greatly aids modern readers, giving the poems a somewhat theatrical feel. In antiquity, in an environment where the audience relied solely on auditory cues, it would have been indispensable to provide clear signals so that listeners could unambiguously follow the dialogues. Therefore, the modern convention of appending the speaker’s name appears anachronistic. It is more likely to be an editorial solution designed to help readers in navigating the complex interchanges of the poem rather than a reflection of Homeric or early performative practice. This led me to wonder how Homer originally indicated (or even encoded) these shifts in speech.

To answer, I turned to the primary text itself. Because I do not read Ancient Greek, I was primarily concerned with the structural aspects rather than lexical detail. In essence, focusing on the “architecture” of speech rather than dissecting its constituent parts. My goal was to determine if the epic contained explicit “delimiters”, or markers distinguishing Homer’s narrative passages from the direct speech of characters.

The idea of delimiters, as outlined in my previous article, posits a general principle: every interpretive language uses markers that signal transitions from a primary layer to an embedded level of expression. In contemporary English prose, quotation marks serve precisely this function; they separate the narrator’s description (e.g., It is raining) from a character’s utterance (e.g., Alice said: “It is raining”). I sought to identify analogous signals within the Homeric text that fulfil the same role.

In Homer’s era, punctuation marks such as quotation marks have not been invited yet, and even if they had existed, they would have been invisible to listeners in a purely oral performance. Therefore, my search was confined to intrinsic textual structures that could signal a shift from narrative to direct speech. It is plausible that a delimiter can be discerned through its structural properties when it shares the same alphabetic system as the surrounding text (in other words, when it is not a special character). As demonstrated in the previous article, languages can employ distinctive patterns, like palindromes, to mark boundaries. Thus, I sought regular linguistic constructs within Homer’s poems that could serve this delimiting function.

I used two resources to access the original text, hosted by Tufts University: the Perseus Digital Library (Greek text and English translation by A.T. Murray) and the Beyond Translation project (Greek text by David B. Munro and Thomas W. Allen). In this article, most of the examples come from the first book of the Iliad and the first book of the Odyssey to keep it readable.

My analysis revealed that Homer frequently employs formulaic constructions as structural cues for navigating between narrative and direct speech. These recurring phrases function as delimiters in three distinct roles:

Opening delimiters signal that the narrator yields the floor to a character (first‑to‑second order transition). 1.

Closing delimiters indicate that a character’s speech has concluded, and the narration resumes (second‑to‑first order transition). 1.

Transition delimiters are used when one speaker responds directly to another within a continuous block of dialogue (second‑to‑second order transition).

Opening Delimiters. The sentences that start a speech are highly variable. For instance: “… and he implored all the Achaeans, but most of all the two sons of Atreus, the marshallers of the people:” (Iliad 1.15-1.16) or “… yet the thing did not please the heart of Agamemnon, son of Atreus, but he sent him away harshly, and laid upon him a stern command:” (Iliad 1.24-1.25), or “… and spoke swift-footed Achilles” (Iliad 1.58). Homeric opening delimiters lack a fixed form, yet they sometimes contain the enigmatic formulaic phrase “ἔπεα πτερόεντα [winged words]”, like in “ἀγχοῦ δ᾽ ἱσταμένη ἔπεα πτερόεντα προσηύδα: [and she drew near to his side and spake to him winged words:]” (Iliad 5.123). As noted by a researcher: “All the usages of epea pteroenta / epea pteroent’—without any exceptions—introduce the direct discourse of a character in the epic.” (Françoise Letoublon. Epea Pteroenta (“Winged Words”). Oral Tradition, 1999, Oral Tradition, 14 (2), pp.321–335, p. 331). While all verses containing ἔπεα πτερόεντα are opening delimiters, all opening delimiters do not necessarily contain this expression.

The variability of opening delimiters does not appear to be surprising. When a bard sang the poem, in a first-order context (as the narrator), they just announced a speaker by preterition. For example: “[Agamemnon] laid upon him a stern command” implicitly announces that Agamemnon is about to speak. The transition to the second-order expression is obvious.

Closing Delimiters. In contrast, signaling a return to narrative from direct speech proves more problematic for an oral reciter. A speaker might employ a phrase such as “to conclude …,” but such strategies are difficult to systematize. To indicate a closure, Homer had to frequently use repetitions such as “ὣς ἔφατ᾽ [so he spoke]” (Iliad 1.33, 1.43), “ὣς φάτο [so he spoke]” (Iliad 1.188, 1.245), “ὣς ἔφαθ᾽ [so spoke]…” (Odyssey 1.43), “ὣς εἰποῦσ᾽ [so she spoke]” (Odyssey 1.95), “ὣς εἰπὼν [so saying]…” (Odyssey 1.125), or “ἤτοι ὅ γ᾽ ὣς εἰπὼν [when he has thus spoken…]” (Iliad 1.69, 1.101). They appear after the speech, much like closing quotation marks follow the enclosed text. By maintaining a uniform structure, these formulae enable listeners to recognize immediately that the current character has finished speaking and that the narrative voice resumes.

Transition Delimiters. As outlined earlier, transition delimiters demarcate the boundary between successive speakers: speaker A delivers an utterance; a delimiter signals that speaker B is now about to speak; then speaker B speaks. These delimiters typically begin with the particle τὸν δ᾽ [and him] or τὴν δ᾽ [and her] followed by the speaker and a verb indicating the act of speaking. For instance: “τὴν δ᾽ αὖ Τηλέμαχος πεπνυμένος ἀντίον ηὔδα: [Then wise Telemachus answered her:]” (Odyssey 1.230) or “τὸν δ᾽ ἀπαμειβόμενος προσέφη πόδας ὠκὺς Ἀχιλλεύς: [In answer to him spoke swift-footed Achilles:]” (Iliad 1.84). As with the closing delimiters (ὣς…), the initial word of a transition clause strongly predicts that a delimiter is present.

What I find interesting with these transition delimiters is their frequent pairing with an adjective that characterizes the speaker: the wise Telemachus, the swift-footed Achiles, etc. For instance: “τὸν δ᾽ ἠμείβετ᾽ ἔπειτα ἄναξ ἀνδρῶν Ἀγαμέμνων: [Then the king of men, Agamemnon, answered him:]*” (Iliad 1.172). I therefore wonder if Homer had to use these adjectives to satisfy the metrical constraints by adding syllables necessary for dactylic hexameter. I believe that Homer fine-tuned the total syllabic count of these delimiters by using these adjectives as fillers (i.e., “to him/her, <adjective-as-filler> <name> replied/answered/etc.:”). That would explain why he continuously and repeatedly uses them across both poems. This is just a hypothesis, but I would love to know what specialists think of this idea of using adjectives as fillers for delimiters.

The role of these formulaic expressions has already attracted scholar attention. A notable example is the observation of Françoise Letoublon: “In an oral epic one has a compelling need for signals of direct discourse, in principle both before and after the reported speech, so that the audience will be aware that the narrating bard is assuming the voice of his characters. And these signals must be clear, perceptible even by a less than attentive audience: they must therefore be regular enough to play the role that iconic marks of quotation (“...”) play for us in the written text. . . . Following direct discourse, the signals for closure—equivalent to closing quotation marks in the typographical tradition—at times include epos, but other formulas are used more often, most frequently with a verb of speaking in the aorist tense.” (Françoise Letoublon. Epea Pteroenta (“Winged Words”). Oral Tradition, 1999, Oral Tradition, 14 (2), pp.321–335, p. 333). In other words, the delimiters function as highly recognizable formulaic markers that cue listeners to a shift from direct to indirect discourse (i.e., first‑order to second‑order expressions) and signal a change of speaker. Letoublon illustrates this point by comparing the delimiters to quotation marks: these formulaic clauses demarcate transitions within the narrative flow.

After having applied this intuition about delimiters acting as a fundamental law of any language to LLMs, I suggest using it as a conceptual tool for analyzing ancient texts. I believe that the Iliad and the Odyssey demonstrate that delimiters can be systematically encoded using ordinary alphabetical forms in a natural language. This observation raises an intriguing possibility: if similar delimiter structures exist in ancient undeciphered scripts, they might provide a foothold for decipherment. However, successfully applying this strategy requires establishing whether a target script encodes higher-order expressions such as dialogues, which is very speculative.