Sources and methodology

Sources and licences

Scholarly works (Barnes, Page, Spurkland, Düwel, Price and others) are cited and paraphrased, never copied. Facts written from general knowledge were checked against at least one such source or a primary text; where we could not settle a point, the page says it is uncertain.

Checking the transliterator against real inscriptions

The transliterator follows rules, and Viking-age carvers did not. To measure how close the rules come, every build runs the Younger Futhark engine on the normalised Old Norse text of each Viking-age inscription on this site (with "R" marked as in the database and the final-r option on) and compares the result word by word with what was actually carved.

On 900 Viking-age inscriptions

We ran the same comparison once on the whole Viking-age part of the Scandinavian Runic-text Database: every inscription where the database's normalised Runic Swedish or Danish text and its transliteration have the same number of words, skipping damaged or uncertain words.

By region (corpus prefix): U 50% of 2,389 words; DR 51% of 977 words; Sö 53% of 701 words; Ög 42% of 473 words; Vg 38% of 297 words. The commonest differences show what a letter-by-letter tool cannot know:

The normalised form ok ("and") was pronounced with a diphthong in East Norse and carved auk; stæin is stin in Danish but stain in Swedish; late carvers used dotted runes (transliterated e, d, g) where older ones did not. The comparison used the database's 2008 release (Rundata 2.5); only the totals and the most frequent words are shown here.

On the inscriptions described on this site

Result this build: 11 of 20 words (55%) come out exactly as carved. The differences are the point of the exercise:

Rök runestone

Small Jelling stone (Gorm's stone)

Not compared: 11 normalised words vs 13 carved.

Large Jelling stone (Harald's stone)

Not compared: 28 normalised words vs 31 carved.

Ramsund carving (Sigurd carving)

Piraeus Lion

Not compared: 7 normalised words vs 10 carved.

Typical reasons for a difference: the carver wrote a regional pronunciation (Danish þusi for þausi), wrote a vowel with an unexpected rune, dropped or kept a nasal, or ran two words together; the normalised forms also reflect editors' choices. The tool picks the most common convention; it cannot know which carver you are imitating.

Two engines, one answer

The transliterator exists twice: in Python, which renders the examples into the pages, and in JavaScript, which runs in your browser. Both read the same rule table. On every build the JavaScript version is run on these test sentences and the build fails if any answer differs:

Checks on every build

• Every rune's glyph is a real Unicode runic letter, and its code point and Unicode name are printed from the Unicode database, not typed by hand.
• Every rune-poem reference points to an existing stanza for that rune; every line-of-descent link between rows points to an existing rune.
• Row sizes are 24, 16 and 33.
• Every internal link on every page resolves.
• Python and JavaScript transliterators agree on the test sentences.

Any failure stops the build and the live site stays as it was.

Photo credits

Photos are downloaded once, re-encoded locally at 1280 and 500 pixels wide, and served from this site.