Treasure Emotion Network

Explore 114 emotion and modifier words from Estonian treasure legends (varandusemuistendid), organized into 13 emotion domains and connected by their discovery relationships.

Domains: fear, joy, sadness, anger, love, surprise, trust, desire, courage, shame, anxiety, supernatural, danger.

Data source: treasure emotion lexicon built via the substitution-test pipeline (per-seed Harris substitution → depth-5 expansion → Claude Opus review → 3-layer corrections).

How Nodes Are Calculated

Each node represents one lemma from the treasure emotion lexicon (114 total).

• Node size is currently uniform — the formula is 3 + hit_rate × 15, but all lemmas in this dataset achieved a hit rate of 1.0, so every node appears at the same size.
• Node colour = emotion domain (13 distinct colours visible in the legend on the left).
• Quality tier (filterable): High (≥5 independent detection methods confirmed the word), Medium (2–4), Low (1). “Sources” = how many of the pipeline’s detection paths independently found this word.
• Depth = substitution chain distance from the nearest seed word. Depth 0 = one of the 26 original seed words; depth 1 = found directly as a substitute for a seed; depth 2+ = found via intermediate substitutions.
• PPMI score = how strongly this word is associated with treasure legends compared to the general runosong corpus (pointwise mutual information). Higher = more treasure-specific.
• Corpus frequency = total occurrences across all 564 treasure legend texts.

How Edges Are Calculated

• Chain edges (solid lines): connect two lemmas that appeared consecutively in the same discovery chain — word B was found as a substitute for word A’s template slot. Edges are undirected (no arrowheads); they indicate co-occurrence in the same discovery path.
• Family edges (dashed lines): connect all family members to the family head when no chain edge already links them. These are structural edges representing semantic grouping within the same reviewed emotion family.
• Edge colour = the domain colour of the family they belong to.

Interaction Guide

• Click a node to see: type (E=emotion / M=modifier), family membership, metrics, corpus contexts (up to 3 excerpts with Estonian + English), family members (clickable), and wordform list (clickable → Emotion Mapping page).
• Drag nodes to rearrange the layout.
• Scroll to zoom, pan by dragging the background.
• Left panel: search by lemma, filter by domain / quality tier / max depth.
• Press Escape to close detail panel or this help overlay.

Domains & Related Pages

This page uses a 13-domain classification specific to the treasure legends substitution pipeline. The Emotion Mapping page uses a broader family-based classification with more granular categories — the wordform links bridge between the two views.

See also: Emotion Vocabulary for the full runosong emotion lexicon (38 families, 61,070 wordforms).

Data Sources

• Treasure legends annotations: 564 texts from archival collections (ERA, EKM, EKRK).
• Emotion lexicon: TREASURE_EMOTION_PIPELINE (Veskis 2025–2027, EKKD-III5 project, funded by Ministry of Education and Science).
• Word families: treasure_word_families.reviewed.json (Claude Opus reviewed, 3-layer corrections applied).