The Plato pilot: the first Greek text in the library, read word by word the way the Arabic texts are, with its readers through the ages. Method: docs/EDITORIAL-METHOD.md.
| Work / author / dates | Plato, Symposium (Συμπόσιον), c. 385–370 BCE; Plato c. 428–348 BCE |
| Part added | 201d–212c: Socrates reports Diotima's teaching, from "καὶ σὲ μέν γε ἤδη ἐάσω" to the knocking at the door (the start of 212d is not included) |
| Reader route | /read/symposium-diotima · about /read/symposium-diotima/about |
| Status | working edition, pilot · reviewed by: — |
1. Witnesses
| Siglum | Witness | Kind | Licence | Location |
|---|---|---|---|---|
| Burnet | J. Burnet, Platonis Opera II (Oxford Classical Texts, 1901; 1910 printing) | standard edition | public domain | as digitised by Perseus |
| P | Perseus Digital Library, canonical-greekLit tlg0059.tlg011.perseus-grc2 (EpiDoc; Stephanus milestones) | digital | CC BY-SA 4.0 | github.com/PerseusDL/canonical-greekLit |
| G | GLAUx corpus, document 0059-011 (A. Keersmaekers): the Perseus text with automatic lemma, morphology and syntax | digital, analysed | CC BY-SA 4.0 | github.com/alekkeersmaekers/glaux |
| Bury | R. G. Bury, The Symposium of Plato (Cambridge 1909): text, apparatus and commentary | edition with commentary | public domain | archive.org cu31924073426151 |
Consulted, not used for the text: Dover (Cambridge 1980) and the Nehamas–Woodruff translation (Hackett 1989) are in copyright and were not used.
2. Base and why
Burnet via Perseus: the standard text, openly licensed, already divided by Stephanus section. GLAUx is built on the same Perseus text, so its tokens carry the Perseus section numbers (div_stephanus_section) and needed no alignment.
3. Collation
- The tokens are GLAUx's, in GLAUx's order. They were checked back against P, token by token, to recover P's paragraph breaks (87 breaks: the turns of the dialogue); every GLAUx token was found in P in order.
- No full collation with Bury was made. Bury's text differs from Burnet in a few places in this range. Two are noted in §4 because they affect the sense; Bury's apparatus lists the rest.
- Script:
scripts/plato/build_symposium.py.
4. Decisions
The Greek text is Burnet's, unchanged. Two places where editors differ are recorded, not adopted:
| Where | Burnet (shown) | Other reading | Note | By |
|---|---|---|---|---|
| 208b | ἀθάνατον δὲ ἄλλῃ ("the immortal [partakes] in another way") | ἀδύνατον δὲ ἄλλῃ ("otherwise it is impossible"), printed by Bury (1909, note ad loc.) | The manuscripts have ἀθάνατον. Bury adopts ἀδύνατον and argues against Stallbaum's defence of the transmitted reading. The AI translation follows Burnet. | AI (noted from Bury) |
| 211c | ἐπαναβασμοῖς (T and the Oxyrhynchus papyrus) | ἐπαναβαθμοῖς (W), printed by Bury | Both are built on ἀναβαίνω; -βασμός is the stepping, -βαθμός the step. | AI (noted from Bury's apparatus) |
| 208a | μνήμην | [μνήμην], bracketed by Bury as interpolated or corrupt | Bury argues that ἀπιούσης must refer to τῆς ἐπιστήμης. | AI (noted from Bury) |
Corrections to the automatic analysis (GLAUx). These are listed on the About page, with links to each word:
- 201d, 211d: Μαντινικῆς / Μαντινικὴ. GLAUx lemmatised these as the place name Μαντινέη; they are the adjective Μαντινικός.
- 202d: GLAUx fused a question mark to the next word (";τί"). It has been split.
- 203b, 203c: ὑὸς. GLAUx gave ὗς, "swine"; the word is the Attic ὑός, "son" (υἱός), as in "the son of Mētis".
- 207c: Διανοῇ. GLAUx took it as a noun; it is the verb διανοέομαι, 2nd singular present ("do you think…?").
GLAUx's other analyses are shown as automatic. It reports 97.2% accuracy on morphology and 98.8% on lemmas, so a few further errors should be expected. For example, 210b δεῖ is tagged imperfect.
Compounds and crasis. GLAUx splits οὔτε into οὔ + -τε, and κἀγώ into κα- + κἀγώ. The reader shows each such pair as one word and gives both analyses in the word panel.
5. Quotations
- 205d "μέγιστός τε καὶ δολερὸς ἔρως" and 208c "καὶ κλέος ἐς τὸν ἀεὶ χρόνον ἀθάνατον καταθέσθαι" are printed in quotation marks by Perseus/Burnet as verse tags; their sources are not identified here.
- 209d: Lycurgus and Solon.
- 208d: Alcestis, Achilles, Codrus.
6. Accents and breathings
The Greek is Burnet's, fully accented (polytonic). Nothing was supplied. GLAUx stores the text in Unicode NFD; the reader shows it in NFC, in GFS Didot.
7. Segmentation and word IDs
- There are 55 passages, one per Stephanus section from 201d to 212c. Stephanus numbers appear in the margin, and each passage is linkable as
#s-201d. - Word IDs are
<section>.<n>, where n is the GLAUx token's position within the section, punctuation included (for example201e.72). IDs stay stable while the GLAUx release stays the same.
8. Analysis layer
- Lemma and morphology come from GLAUx. The tags follow the AGDT scheme and are decoded in the reader by
src/lib/greek.ts. - The dictionary is LSJ (Logeion's edited Perseus LSJ, CC BY-SA 4.0), matched by headword. 669 of 682 lemmas were matched. The 13 unmatched are mostly proper names.
- When LSJ has several homographs, the entry that cites Plato most is taken: λέγω "B", ἐράω "A".
- The "opening senses" are the first English gloss of each sense, extracted automatically, and they are marked as such.
- Etymology comes from Wiktionary, via kaikki.org's Wiktextract (CC BY-SA 4.0).
- 597 of 682 lemmas have an entry. 252 of them give Proto-Indo-European forms, which the reader shows as reconstructions.
- Beekes (2010), Chantraine (DELG) and Frisk (GEW) are the standard etymological dictionaries. All are in copyright, so the panel only cites them ("see Beekes s.v. …").
- Boisacq (1916) is the primary etymology (since round 2). É. Boisacq, Dictionnaire étymologique de la langue grecque (Heidelberg/Paris 1916), public domain, archive.org
dictionnairety00bois.- OCR: macOS Vision has no Greek or Latin model (tested on macOS 26.6 with (private copy): it reads the French well and turns the Greek into Latin-letter noise). So all 1,162 pages were OCR'd with Tesseract, tessdata_best
grc+fra(Apache-2.0), at 300 dpi:scripts/plato/boisacq_ocr.sh→data/raw/plato/boisacq/ocr/. - Matching:
scripts/plato/boisacq_index.pyfinds entries by the dictionary's alphabetical order (headword candidates that break the order are dropped), then matches lemmas accents aside. Tiers: the lemma's own entry (286 lemmas); a line or paragraph the lemma opens; the entry that treats it (36); or, when LSJ derives the lemma from a base word, Boisacq's entry for that base, labelled as such (33). A short entry that only says "Voy. s. v. X" also shows X's entry. 355 of 682 lemmas are found; the rest are mostly proper names, compounds with prepositions and particles. - Printed page = PDF page − 34 (681 running heads read, the offset constant through the book); archive.org leaf n = PDF page − 1. The panel shows the machine-read entry marked "OCR, unchecked", links the page, and can show the page image inline.
- Known weaknesses: OCR noise in small Greek; occasionally the matched paragraph belongs to a neighbouring entry. The page image is authoritative.
- OCR: macOS Vision has no Greek or Latin model (tested on macOS 26.6 with (private copy): it reads the French well and turns the Greek into Latin-letter noise). So all 1,162 pages were OCR'd with Tesseract, tessdata_best
- LSJ's own etymological remarks: the derivation LSJ gives after the headword ("τόκος (τίκτω)", "ἔρως (ἔραμαι, ἐράω A)") and its occasional cognate notes ("Cf. Skt. …"), extracted automatically for 193 lemmas (
scripts/plato/lsj_etym.py). LSJ's<etym>tags were checked and are too inconsistent to use. - Wiktionary is kept as a secondary, labelled source (collapsed when Boisacq has the word).
- Word families are AI groupings of the pilot's lemmas by stem (for example ἐρ-: ἔρως, ἐράω, ἐραστής, ἐραστός, ἐρωτικός), made from the etymologies and marked AI.
- Usage lists each lemma's occurrences in this passage and in the Phaedrus (GLAUx
0059-012, linked to Scaife).
9. Commentaries aligned: "Readers through the ages"
- Data:
app/src/content/plato/symposium-diotima-braid.json, 20 entries. - Alignment: each entry is keyed to the Stephanus sections it discusses. It is marked exact when the reader quotes or paraphrases those lines, and approximate when the discussion ranges more widely. The reader shows an entry under the first section it covers and points to it from the others.
- Originals: all public domain. Latin and Greek excerpts were cleaned from the archive.org OCR by AI (long s, ligatures, misread letters such as ϑ, and dropped initials, which are shown in brackets). They still need checking against the linked page images. Their English is AI and marked.
| Reader | Where | Sections | Source |
|---|---|---|---|
| Plotinus, Enn. III.5.2 and III.5.9 | Greek of III.5.9 (Volkmann 1883); MacKenna's English (1921) for III.5.2 | 203b–e | archive.org volkmann-enneades-gr-v.-1-1883, plotinustranslat00plot |
| Hermias, In Phaedrum | Couvreur pp. 12, 66, 72: Eros as a great daimōn; the daimonic kind; Eros as philosopher, enchanter, sorcerer, sophist | 202d–203a, 203d | archive.org hermiaealexandr00couvgoog |
| Ficino, De amore VI.3, 7, 9, the chapter on love's use, 15, 18, 19 | Latin, Lyon 1590, pp. 784–791 | 202e–203d, 206e–208b, 210a, 210e–212a | archive.org diviniplatonisop00plat |
| Taylor (1804) | notes signed "T." on the great daemon, and on Plenty, the nectar, the garden and sleep | 202d–203c | archive.org vol3worksofplato00plat |
| Sydenham | note on the mysteries (in Taylor's edition) | 209e–210a | as above |
| Bury (1909) | notes on τόκος ἐν καλῷ, κυοῦσι and the 208b crux; apparatus on ἐπαναβασμοῖς | 206b, 206c, 208b, 211c | archive.org cu31924073426151 |
- Hermias: he wrote on the Phaedrus, not the Symposium, so he appears only where he discusses the Symposium directly.
- Round 2: 24 more readers (
app/src/content/plato/symposium-diotima-braid-more.json, built byscripts/plato/build_braid_more.py), all public domain and each linked to its page scan:
| Reader | Where | Sections | Source |
|---|---|---|---|
| Plotinus, Enn. I.6 "On Beauty" | I.6.1, 7, 8, 9: Greek (Volkmann 1883) with MacKenna's English (1917) | 210a–212a | archive.org volkmann-enneades-gr-v.-1-1883, plotinustranslat01burkuoft |
| Proclus, Theology of Plato I.23, 24, 26, 27 | Taylor's English (1816), from the 1816 scan, not the 2010 e-text | 202d–204c | archive.org sixbooksproclus00unkngoog |
| Leone Ebreo, Dialoghi d'amore III | Rome: Blado 1535, fols. 95r–96r; Italian with AI English | 203b–204b | archive.org b33349149 (Wellcome) |
| Pico, Commento (tr. Stanley 1651, ed. Gardner 1914) | Book II ch. 10; Book III, the six degrees | 203b–c, 210a–211c | archive.org platonickdiscour00picouoft |
| Castiglione, Courtier IV (tr. Hoby 1561) | Bembo's stair of love; the heavenly beauty; Giuliano naming Diotima | 201d, 210a–211b | archive.org bookofcourtierfr00castuoft (Tudor Translations 1900) |
| Jowett's Introduction (1892) | on Socrates and Diotima; on her dialectic; on the ladder | 201d–e, 204b, 211c | archive.org dialoguesofplat01plat |
| Bury's Introduction (1909) | Diotima and Mantinea; Penia; the garden; the ebb and flow of Eros; the ladder as dialectic | 201d–210c | archive.org cu31924073426151 |
- Checked and left out: Shelley's "Discourse on the Manners of the Ancient Greeks Relative to the Subject of Love". The 1840 fragment breaks off before the Symposium itself, and the fuller 1931 Ingpen printing is not public domain in the US until 2027. Proclus's In Timaeum and In Alcibiadem were not needed.
9a. Bridges to the Arabic library
- Data:
app/src/content/plato/bridges.json, 6 bridges, each AI-labelled with a confidence and its grounds (primary texts linked, studies cited only):- καλόν ↔ ḥusn / jamāl (high)
- ἔρως ↔ ʿishq / ḥubb (high)
- ἐπαναβασμοί ↔ miʿrāj / taraqqī (medium)
- δαίμων, μεταξύ ↔ barzakh (medium)
- τόκος ἐν καλῷ ↔ nikāḥ / wilāda (low)
- μανία ↔ jadhb / hayamān (low; for the Phaedrus)
- In the reader: the word panel shows a Bridges tab on words of those lemmas, and ◆ marks the first such word in each section. The About page lists all six.
- Grounding:
- Ibn Sīnā, Risāla fī māhiyyat al-ʿishq: OpenITI lines 115 and 120–128.
- The Ikhwān epistle 37: OpenITI lines 8924–9069; "the divine sickness" at line 8930.
- Dieterici's 1882 Arabic Theology of Aristotle.
- Futūḥāt chs. 63, 167 and 178 (cited by chapter; not checked line by line in this session).
- Caution: a bridge names shared vocabulary or structure, never borrowing, unless a source shows it.
10. Translation
- Jowett, The Dialogues of Plato, 3rd ed. (Oxford 1892), vol. 1, pp. 572–582. Public domain.
- The text is Project Gutenberg #1600, collated automatically against the 1892 scan (archive.org
dialoguesofplat01plat). - Two readings where Gutenberg departs from the 1892 edition were restored to 1892: "I will rehearse" (201d), and "run risks of all kinds greater far" (208c).
- Jowett's cross-references, which Gutenberg prints inline, were removed.
- Jowett has no Stephanus numbers. His text was divided into sections automatically, by the best monotonic match of each clause against Lamb's and the AI English for each section, so the boundaries are approximate. 210c received no clause of its own; Jowett's rendering of it falls in 210b and 210d. Script:
scripts/plato/align_jowett.py.
- The text is Project Gutenberg #1600, collated automatically against the 1892 scan (archive.org
- Lamb, Loeb vol. 3 (1925), from Perseus
tlg0059.tlg011.perseus-eng2, divided by Perseus's own section milestones.- Public domain in the US.
- Lamb died in 1961, so outside the US the translation may still be in copyright. The Perseus digitisation is CC BY-SA 4.0.
- Lamb's 201d milestone falls inside Socrates' reply to Agathon, so his 201d begins at "And now I shall let you alone", where the Greek 201d begins.
- AI English (Claude), written for this reader from Burnet's Greek, without consulting any copyrighted translation, and marked AI.
- Fixed renderings: ἔρως "love" (the god: Love); καλόν "beautiful"; δαίμων "daimōn"; τόκος ἐν καλῷ "giving birth in beauty"; ἀρετή "excellence"; φρόνησις "understanding"; ποίησις "making" (and "poetry").
10a. Further translations, for comparison (round 2)
Each is a file in app/public/data/plato/tr/, fetched only when chosen (Translations or Compare in the reader); each file records its source, licence, how its text was made and how it was divided. scripts/plato/publish_translations.py checks every file has all 55 sections.
| Translation | Printing used | Text | Sections | Licence |
|---|---|---|---|---|
| Ficino, Latin (1484) | Lyon 1590 Opera, pp. 327–331 | transcribed by AI from the page images (the OCR interleaves the Greek column) | by AI reading against the Greek | PD |
| Sydenham (1767) as revised by Taylor (1804) | Taylor, Works of Plato III, pp. 494–516 | OCR, long s restored against Sydenham's 1767 edition, ~200 errors fixed on the scans | automatic, then 34 of 54 breaks moved by hand | PD. Taylor revised Sydenham (listed in the file); the knocking at the end of 212c is not in Sydenham, who stopped at Socrates' speech |
| Schleiermacher (1807) | Platons Werke II.2, pp. 417–434 (BSB scan) | transcribed by AI from the scan, checked against the library OCR | by AI reading, checked against the print's marginal Stephanus pages | PD |
| Shelley (1818; publ. 1840) | Mary Shelley's 1840 Essays, vol. 1, pp. 126–148 | two OCRs merged and corrected against every page | automatic, then 31 breaks moved | PD. The 1840 text is softened in places (noted); the unexpurgated text (1930/31) is not used |
| Cousin (1831) | Œuvres de Platon VI (1849 reissue scan) | French Wikisource's proofread text | Wikisource's anchors corrected (10 wrong), by AI reading | PD; transcription CC BY-SA |
| Burges (1850) | Bohn, Works of Plato III, pp. 531–556 | OCR cleaned; his footnotes and section numbers removed | automatic, then 36 breaks moved | PD |
| Chambry (1919) | Garnier 1919, pp. 385–402 (Gallica) | transcribed by AI from the scan (the Wikisource text is the revised 1964 reprint) | by AI reading | PD in the US; in the EU since 2022 (Chambry d. 1951) |
| Allen (1991), private | Yale UP, via JSTOR (USD library) | (private copy) only (gitignored), footnotes omitted; shown only when PRIVATE_LAYERS=1 | by Allen's own margin letters (43 of 55), the rest by overlap with Lamb and the AI English | © Yale. A public build was checked: it contains none of it |
11. Rights and private layers
- Everything shown is public domain or openly licensed (see the About page).
- Share-alike layers: Perseus, GLAUx, LSJ and Wiktionary are CC BY-SA 4.0, so derived data files, such as
symposium-diotima.jsonand the lexicon, carry the same licence. - Private or unused: Nussbaum (Fragility of Goodness), Carson (Eros the Bittersweet) and the Pierre Grimes transcripts are in copyright or unclear, and are not used. The same goes for Steadman's Symposium (CC BY-NC-SA 3.0): it is non-commercial and its glosses were not used, so as to keep the public layer free of NC terms.
- Audio: none, but the Arabic readers' background music is available. Azure's Modern Greek voices would cost about $0.33–0.36 for the whole passage; no spend was made. See
docs/research/greek-audio-options.md.
12. Open questions / to review
- Audio. Choose between a Modern Greek neural voice with the Erasmian/Modern pronunciation question stated, a human reading, or no audio.
- Review. A person should check the AI translation and the AI glosses, especially the particles.
- Bury. Should Bury's readings be shown as a second text, with a word-by-word collation, or only in notes as now?
- Jowett's section breaks. The automatic division could be checked by hand in about an hour.
- The Phaedrus. The palinode (243e–257b) can follow the same pipeline: the GLAUx and Perseus files are already acquired.
13. Change log
- 2026-09-28: pilot built. Greek, lemmas, LSJ, Wiktionary, Jowett, Lamb and AI English.
- 2026-09-28: the readers through the ages (20 entries) and the bridges (6) added.
- 2026-09-28 (round 2): seven further public-domain translations with a picker and a compare mode; Allen as a private layer; 24 more readers; Boisacq as the primary etymology; background music; audio options researched (
docs/research/greek-audio-options.md).