What is this page?
This explorer maps the emotional landscape of Finnic runosong across 803 collection places in Estonia and Finland. Each place is scored by how often emotion-bearing words appear in its poems, based on DeepSeek AI analysis of 165,000 poems. The result is a geographic view of where different feelings concentrate in oral tradition.
Emotion domains and families
Emotions are organized into
26 domains (e.g., Love, Grief, Anger, Fear) that group
38 finer-grained families. For example, the Grief domain contains families like mourning, weeping, and bereavement. The left panel lists every domain; click the arrow to expand and see individual families. Selecting a domain or family recolors the map to show only that emotion's distribution.
The choropleth map
Each circle on the map represents one collection place.
Circle size and opacity encode the value: larger, darker circles indicate higher concentrations. The color matches the selected domain or family. In domain or family mode, circles reflect
total occurrences of that emotion's words at each place (raw token count). In "Overall Emotional Density" mode, circles show a
density ratio: the number of emotion word occurrences per 10,000 total words at that place (repeated words count each time they appear), which normalizes by word count.
Normalization: per 1000 poems
In domain or family mode, the map shows raw occurrence counts by default. Enable
"Per 1000 poems" to normalize by poem count instead, dividing by the number of poems at each place. This corrects both for uneven collection sizes and for the effect of long poems with repeated emotion words (e.g. in refrains), since each poem counts equally regardless of length. The toggle is disabled in density mode, which already normalizes by word count.
Minimum poems filter
The
Min poems slider hides places with too few poems, which tend to produce noisy or unreliable percentages. A threshold of 20-50 is recommended for meaningful patterns.
Language filter
Use the
ET / FI / All buttons to show only Estonian (ERAB) or Finnish (SKVR + JR) places, or both. This is useful for comparing emotional patterns between the two traditions.
Place detail and radar charts
Click any circle to open the right panel with that place's full emotion profile. The
radar chart shows the place's score across all 25 emotion domains (excluding "Other"), normalized against the global maximum for each domain. A place that fills a particular axis completely has the highest count in the entire corpus for that emotion. Below the radar chart, a
bar chart lists the top 12 emotions ranked by raw count.
Comparing two places
Shift+click a second place to overlay its radar polygon (shown in pink dashed line) on top of the first. This makes it easy to compare the emotional signature of, say, an Estonian island parish with a Finnish inland area.
Place search
Type at least two characters in the
Find place field to jump to the first matching location and open its detail panel.
Statistics section
Below the map, three cards provide corpus-wide summaries:
- Emotion Hotspots — the top 15 places for the currently selected emotion (updates when you switch domain/family).
- Estonian vs Finnish — side-by-side comparison of emotion rates per 1000 poems in each tradition, revealing which emotions are more prominent in Estonian vs. Finnish runosong.
- Emotional Diversity — places ranked by Shannon entropy of their emotion distribution (minimum 50 poems). High entropy means a wide, even spread of emotions; low entropy means one or two emotions dominate.
Why some families appear only in Estonia or Finland
The emotion vocabulary is built by combining 12 detection methods (see the
Emotion Vocabulary browser for the full methodology). One of those methods — the
substitution test, which discovers words that fill the same slot in Estonian verse templates — is Estonian-internal by construction and naturally finds more Estonian candidates. The translation-based methods (SetFit, GoEmotions, DepecheMood++, NRC EmoLex) cover both corpora through DeepSeek R1 English translations of all 1.2M word forms. Some families (e.g.,
Hardship, anchored on Estonian
raske) were discovered primarily through Estonian-internal methods, while their Finnish semantic equivalents were classified into neighbouring families (e.g.,
Pain or
Lament). When a family shows no or few bubbles in one country, it usually reflects a classification gap rather than the absence of that emotion in the tradition.
Data source
Emotion annotations come from the RunoVerse emotion lexicon (38 families, 63,871 word forms, built by consensus over 12 detection methods and reviewed by Claude Opus), aggregated at the collection-place level. Place coordinates are from ERAB, SKVR, and JR metadata. The underlying index (
emotion_place_index.json) contains only place names, numeric counts, and pre-defined domain/family identifiers. For the full pipeline and the list of emotion wordforms behind each family, see the
Emotion Vocabulary browser; for the poem-level view, see
Emotion Canvas.