Emotion pipeline results for 6,195 Estonian treasure legends — RunoVerse
Treemap of 13 canonical emotion domains sized by wordform count.
Top 30 P-types with strongest emotion signals (total annotated tokens across all domains).
Top 20 parishes by emotion word density (emotion words per 1,000 tokens).
How the two detection systems overlap in discovered emotion words.
E-word counts by discovery depth (System A expansion rounds), coloured by domain.
Distribution of hit_rate (fraction of PPMI sources confirming the word) across all System A words.
Domain distribution overlap between runosong poetry and treasure prose emotion lexicons.
Top 50 emotion words ranked by corpus frequency × confidence score.
How often pairs of emotion domains appear in the same record (out of 6,195 records).
Average position (quartile 1–4) of emotion words within each record. Q1 = beginning, Q4 = end.
Discovered E-words: PPMI score (distance from nearest seed) vs hit_rate. Higher PPMI = closer to seeds.
Each of the 93 treasure-legend types carries a distinct emotional signature. This explorer normalises the raw emotion signals (mentions, not poems) so large and small types are comparable — in the default view every row sums to 100 %. Switch to Over/under-index to see how far a type departs from the corpus average, the clearest measure of what makes a legend type emotionally distinctive. Distinctiveness is most reliable above the record floor (default ≥ 40); raw counts and corpus comparisons appear on hover. Data: treasure_emotion_by_type.json (loaded live).