Song Type Explorer

About Song Types

Runosong types are scholarly classifications of poems by their content and function. The Finnish SKVR corpus assigns type labels (runotyypit) to 98.6% of its poems, and the Estonian ERAB corpus has type labels (laulutüübid) for 65.5% of its poems. Types include incantations, lullabies, epic narratives, wedding songs, and many more. The English names are AI contextual translations of the original type names, with manual corrections.

How to Use

Search for types by name or English translation. Filter by language (Finnish/Estonian) and theme category. Click a type to see its geographic map, temporal distribution, co-occurring types, and sample poems. Sort by frequency, alphabetically, or by geographic concentration.

Geographic Concentration (HHI)

The Herfindahl-Hirschman Index measures how concentrated a type is geographically. Low values (near 0) mean the type is widespread across many places; higher values mean it was collected primarily in a few areas.

END Association

END (Emotion Neighbourhood Density) measures how much a song type's vocabulary co-occurs with emotion lexicon items across the corpus. A high END score means the type's characteristic words tend to appear in poems that also contain many emotion words. This does not directly measure emotional intensity — incantation and healing types may score high because their vocabulary (body parts, ailments) overlaps with the emotion word list. Scores are language-adjusted (compared against same-language baselines) so Finnish and Estonian types are comparable.

Data Source

Song type labels come from SKVR metadata (Finnish) and ERAB metadata (Estonian). Types are shown in their original language; the English names are AI contextual translations of those names, with manual corrections.

The type names and classifications come from the archival SKVR and ERAB catalogues; the English type names are AI translations (with manual corrections).

List and Tree Views

Use the List / Tree toggle to switch how types are shown. List is a flat, sortable view of every matching type. Tree arranges types under the original archive catalogue hierarchy (the SKVR, ERAB and JR typologies), so you can browse by family rather than by name. Your language, theme and search filters apply to both views. Tree view is available in the All, Finnish and Estonian lists; it is turned off for type groups.

Clusters View

The Clusters view — the one you land on when you open this page — is a zoomable bubble map that arranges the song types into 111 broad families and 9,644 finer groups of near-identical or closely related songs. Of these groups, 1,828 hold two or more song types; each of the others holds a single type that has links to other types but was not grouped with any of them. Each large bubble is a family; the smaller bubbles nested inside it are its groups. Click a bubble to zoom in, click a group to list the individual types it holds, and click any type to open its full detail. Captions sit centred on the bubbles roomy enough to hold them; smaller bubbles stay uncaptioned to keep the map readable, but hovering over any bubble shows its name and how many types it contains. To pinpoint a particular cluster — including the many too small for a caption — use the Find a family box beside the map: type or pick a name and its bubble flashes and zooms into view. Clicking an on-map caption flashes its bubble too, so the link between a name and its circle is always unmistakable. The Labels switch reads the captions in English, Estonian or Finnish. Apart from three ERAB types, each known from a single song that is not in Estonian, no group joins a song type known only from ERAB with one known only from SKVR, KR or JR. When the same song is filed under one type in ERAB and another in SKVR, KR or JR, the group of that other type is often among the nearby groups listed when you open a group. The filter buttons above the map work here too: they light up the groups holding at least one song type that matches the choice, and the families those groups belong to. The map leaves out 1,842 song types, because none of them is linked strongly enough to any other type; most are known from a single poem. They can still be found in the List view.

When a filter is on, a group in which only some song types match the choice is drawn pale. Its bubble fills from the bottom like a glass: the height of the fill shows what share of the group’s types match. A thin band at the bottom of a bubble means that only one or two of many types match; a group in which every type matches stays fully coloured. The list beside the map then starts with the families (and, inside a family, the groups) that hold the most matching song types. The number after each name still counts all their types.

How the clusters were built

The grouping is computed, not hand-curated. For every pair of song types we combine five signals — similar poems, shared verses, poems filed under both types, verses shared across the corpora (the Song Bridge, counted only when the two types share at least three verses), and shared semantic category — into a single connection strength, and keep the connections that reach a set minimum (135,555 of them) as a network. For a link between a type whose songs are all in Estonian and one with no song in Estonian, the minimum is higher, and the shared semantic category does not count towards it. The English translations of the type names are not used. A community-detection algorithm (Leiden) then finds the tightly connected groups, and bundles those into the 111 broad families. While the groups are found, links between a type whose songs are all in Estonian and one with no song in Estonian are set aside, so no group joins two such types; these links do count when the groups are bundled into families and when the nearby groups are chosen. Finally, each new group of two or more types was named in English, Estonian and Finnish by an AI language model, which was shown up to 40 of the largest song types in the group, their English names and the strongest links within the group; groups unchanged from the previous map keep their earlier names. Each family was named the same way, from the names of its largest groups and song types. The grouping is a fresh computational reading of the corpus, so it complements — rather than replaces — the original archive typologies in the Tree view.

Theme Distribution

The bar chart above the list shows how many poems fall under each theme category. Click a bar to filter the list down to that theme.

The theme comes from the archives’ filing of the type or its songs where there is one, otherwise from words in the type’s name.

Semantic Categories

In a type's detail view, this section shows which "semantic worlds" the type's vocabulary belongs to — for example body parts, nature, or kinship — based on an analysis of the words that recur in its poems. Each category links through to a fuller comparison, and the sample words link to the word lexicon.

Genre Cross-Reference

For Estonian (ERAB) types, the detail view can show the archive's own genre classifications (genre, song category and function) for the poems in that type. Only ERAB poems carry this metadata, so the section does not appear for Finnish-only types.

Keyboard Shortcuts

/ — Focus search box
j or ↓ — Select next item in list
k or ↑ — Select previous item in list
Enter — Open selected item
r — Pick random type from filtered list
s — Share URL (copy to clipboard)
x — Export CSV
? — Toggle this help panel
Escape — Clear search / close help