This reader shows AI-generated word-level annotations for runosong texts. The annotations are approximate and intended as a reading aid, not for scholarly citation without manual review.

Layers: Toggle annotation rows using the buttons above the poem. Each poem is shown in its original published spelling (capitalization and punctuation), and every word stays clickable. All layers start switched off. Translation adds an English translation of each verse line, made by the DeepSeek R1 AI model, and Word by word gives the English meaning of each word taken on its own. Commentary adds an AI-generated philological commentary on each verse (DeepSeek R1). Press Normalized to switch the word grid to its lowercased lookup form and reveal the standard-orthography row used for annotation.

Search: Use the search field in the header to jump to poems by archive reference (e.g., “H II 1, 389”) or search for words/lemmas to find all poems containing them.

Click any word to see its full entry including translations, morphology, etymology, and dictionary definitions.

Related Poems: The sidebar shows poems similar to the current one, found by five algorithms: TF-IDF (lemma), Jaccard (wordform), Translation-pivot (cross-corpus), Verse-aggregated RRF, and Character Bigram. Click any result to navigate to that poem.

Verse Similarity: Click the expand arrow (↓) on any verse line to see its most similar verses from across the corpus, with algorithm badges (J/T/Tr/C/Se) and cross-corpus indicators. A small blue dot on the corner of an expand arrow marks a verse whose top matches include a parallel from the other corpus group (Estonian vs. the Finnish-side collections). The verse-map “N verses with parallels in ...” stat is clickable: it highlights those verses, and each highlighted verse has an ET↔FI chip that opens its closest matches at a cross-corpus parallel. Detection uses each verse's stored top matches (up to five per algorithm), so the count is not exhaustive. The Finnish-side corpora (SKVR, KR, and JR) include Karelian, Izhorian, and Ingrian material, so the label describes corpus groups rather than identifying the language of every verse. The site's dedicated Song Bridge and cross-corpus cluster browser use cluster-level analysis and therefore show different counts.

Poem Profile: The dashboard below the poem analyzes the current text (or any text you paste via Paste text). Hover any card in the statistics row for its definition: besides verse and word counts, unique words counts distinct word forms, glossed is the share of those with an English gloss, emotion the share of words from the emotion lexicon, parallel the number of detected parallelism groups (consecutive verses restating the same idea), recurring the average verse recurrence (0–100), and morph the share of the 65 grammatical categories the poem's word forms use. The tabs go deeper: Verse Map shows per-verse rarity, recurrence, parallelism and cluster marks, plus how many poems share each verse — the RV column from RunoVerse's own clustering (click a number to open that verse's similar verses) and the CW column from the FILTER project's clusters. Semantics covers emotions, the words that stand out and Estonian–Finnish cognates; Sound & Meter alliteration, syllable meter and vocabulary rarity; Tradition song types, similar poems, substitution partners (words attested taking a given word's place in otherwise near-identical verses, with a corpus-wide count) and verse clusters; Words the poem's vocabulary and compounds; Context the collector, place and collection period. Throughout the profile, hover any label, count or badge for a plain-language explanation.

Word Detail: Click any word to see its full entry including translations, morphology, etymology, dictionary definitions, and dialectal synonyms. Links to the main lexicon for deeper exploration.

Comparison: Compare two poems side by side with shared vocabulary highlighting. Available via the compare button in the Related Poems panel.

Keyboard: Use arrow keys to move between words, Enter to open in lexicon, Esc to close panels. Press / to search, ? for help, K for the full keyboard shortcuts reference, 1–7 to toggle layers (1 = Normalized, incl. alliteration), [ / ] for prev/next poem, { / } for browsing history back/forward.

Metadata links: In the line under the poem title, the place opens that place's profile page, the year opens the corpus timeline at its decade, and the collector's name opens everything that collector recorded.

Suggested type: Where the archive records no song type for a poem, the line under the title may show a song type suggested by RunoVerse, in grey. It comes from the song types the archives gave to the poems most similar to this one: a language model read the poem and chose among the types those poems carry most. Only the suggestions most likely to be right are shown, as judged by blind checks in which two other language models tested a sample. Where the archive does record a type and RunoVerse would choose a different one, that choice may follow the archive's types, in grey, marked “Also suggested”; these are held to a stricter standard. It may be another name for the same song, or a different song. A suggestion is not an archive classification: song type pages, maps and counts elsewhere on the site use only the archives' own types.

About the corpora

The Finnish corpora (SKVR, KR, JR) also hold Karelian, Izhorian and Ingrian material, so this filter groups collections rather than identifying the language of each poem.

Sort options

Recommended — balanced by verse count with collection diversity. Longer poems rank higher.

Most unique — poems unlike anything else in the corpus, based on vocabulary, structure, and cross-corpus comparison (uniqueness percentile).

Most traditional — poems at the heart of widely shared traditions, with the most variants across the corpus.

Most parallel — ranked by consecutive parallelism score (only poems with computed scores shown).

Random — seeded shuffle, shareable via URL.

About themes

The groups sort the songs of all four corpora by kind. Where SKVR or ERAB has filed a song or its song type under a matching category, that filing decides the group (for some song types, the category is the one where most of their songs are filed). Otherwise the group is worked out automatically, from the song type's name or its similarity to other song types, and the card marks the group “inferred”. Literary Epics holds the epic editions and their manuscripts. A song can be in more than one group.

The coloured stripe on the left edge of a card is its theme colour; cards from poems with no theme keep a grey stripe.

A dashed stripe means the theme was inferred from closely similar poems rather than read from a song type classification.