And what they are not. Every column on these two books names its source, because a translation you cannot trace is a translation you cannot check.
Both books put the original on the left and one living language on the right. The originals and the historical translations are in the public domain. The default reading layer is a new 2026 model translation — declared, dated and traceable, see The new translations below. There is no paywall and nothing to sign up for.
Two medieval Icelandic books: the Poetic Edda (the Codex Regius poems, copied in Iceland around 1270) and Snorri Sturluson's Prose Edda, written around 1220. How old the poems themselves are is still argued — the manuscript date and the poem date are not the same thing.
| Column | Source | Covers |
|---|---|---|
| Old Norse | Guðni Jónsson's normalised text | 38 works, 1,615 numbered units |
| English | Henry Adams Bellows, 1923 (poems) Arthur Gilchrist Brodeur, 1916 (Gylfaginning) |
all works |
| Svenska | Erik Brate, Eddan, 1913 | the Poetic Edda — 32 of 38 works. Brate did not translate the Prose Edda, and the four works added in August 2026 are not paired with him here. |
| Norsk | Ivar Mortensson-Egnund, Edda-kvæde (Gudekvæde), 1905 — nynorsk | the mythological poems — 11 of 38 works. His heroic volume is not yet paired, so the Sigurd and Gudrun poems have no historical Norwegian here. |
| Dansk | House translation | 34 of 38 works — not a named scholarly edition |
| Íslenska | House text — the Old Norse in modern Icelandic orthography | 34 of 38 works |
| New translation | Claude (language model), August 2026 — declared below | all 38 works, five languages — the default reading layer |
One anonymous Old English poem, surviving in a single late-tenth- or early-eleventh-century book (the Nowell Codex, British Library Cotton MS Vitellius A.xv). It has no medieval title. It is not a Viking epic, and it is not England's Homer.
| Column | Source | Covers |
|---|---|---|
| Old English | James A. Harrison & Robert Sharp, 4th ed. | 43 fitts, 3,184 lines |
| English | Francis Barton Gummere, 1909 | 43 fitts, 3,114 lines |
| Nordic (historical layer) | House study — Dansk, Norsk (bokmål), Svenska, Íslenska | the opening 3 fitts only, on the historical layer; the picker says so there. |
| New translation | Claude (language model), August 2026 — declared below | all 43 fitts in five languages — the default reading layer |
The picker says Norsk on both books, but they are not the same Norwegian. The Eddas carry nynorsk, because that is the register Mortensson-Egnund wrote in and his is the public-domain Norwegian Edda. Beowulf's house Nordic is bokmål. Naming the register on a button would be more precise and less readable; naming it here seemed the better trade.
From August 2026 these books carry a third kind of column: a new translation, made with a language model (Claude, August 2026), translated from the original text with the historical English as a sense-check only. Where a work has one, it is the text you read by default — contemporary language in all five columns — and the “New translation / Historical editions” control in the corner of the open book switches the whole book between the two layers. The historical editions never left; they are one press away.
Every new translation names its model, its date and its base text in the book's own manifest, and the build refuses a translated file that lacks that record. The rule of this page has not changed: every column names its source — what changed is that a source can now be a translator that is not a person. Where the reader has chosen the historical layer and a passage has no historical translation, the new translation stands in and the paragraph says so; nothing is ever quietly substituted.
Where a column says a translator's name and a year, that is a published edition out of copyright, reproduced as its editor set it. Where it says house translation, it was made for this book and carries no scholarly authority — read it as a help, not as a source to quote.
The scholarly facing English for the Eddas would be Faulkes; for Beowulf, Liuzza, over Klaeber's critical text. None of those are public domain, so none of them are here. That is the whole reason Bellows and Gummere are the ones you are reading.
The picker does not only swap the verse. The foreword, the running heads, the table of contents, the section headings and the interface all follow it. Until 19 August 2026 they did not — you could press Dansk on the opening page and nothing would change for ten pages, because the front of the book was English at every setting.
Two different things are being translated, and they are not equal. The poem titles in the contents are the titles the named edition itself prints — Brate's Valans spådom for Völuspá, Mortensson-Egnund's Voluspaa — harvested, never written here. The foreword and the interface are the publisher's own words, translated; that is publishing a foreword in five languages, not manufacturing a scholarly translation, and it is the only reason it is allowed at all.
Where a language has no title for a poem — Danish and Icelandic have no named public-domain Edda here — the contents row stays English and is dimmed and italic, so you can see at a glance which part of the page is speaking your language and which is not.
It says so. Choose a language with no text for the passage you are on and the button dims, the paragraph is marked, and you get the English with a note that it is the English. An empty column is honest; a column quietly filled with something else is not.
The original keeps its own lineation and so does the translation. They are paired passage to passage, not line to line — Bellows and Gummere both reorder and compress, so a strict line-for-line promise drifts and then quietly lies about which line means which. Where a line is too long for the column it wraps with a hanging indent, so a wrapped remnant never reads as a new line of the poem.
Bellows prints a gap in the manuscript as a row of spaced dots. Those survive here as an ellipsis: something is missing there, and it was missing in 1923 too.
Each plate hangs under the passage it illustrates, not at the front as a poster. They are made images, not photographs of artefacts, and they illustrate one moment each. The Illustrations index inside each book jumps to every one of them.