Conceptual Metaphor Explorer

What is this?

Conceptual Metaphor Theory (Lakoff & Johnson, 1980) shows that metaphor is a fundamental cognitive mechanism — we understand abstract concepts through concrete experience. This explorer is the first systematic computational mapping between an established conceptual metaphor database and folk poetry corpora. The core insight: runosong parallel verse substitution IS conceptual metaphor — when “tamm” (oak) substitutes for “vanamees” (old man) in the same verse position, this directly instantiates PEOPLE ARE TREES.

Data sources

MetaNet (Dodge et al. 2015, Stickles et al. 2016): 875 English conceptual metaphors with formal source/target frame mappings, 951 frames, 9,000 lexical units, from Berkeley ICSI and UBC.
RunoVerse substitution pairs: 701,350 word substitution pairs extracted from parallel verses across Estonian (ERAB) and Finnish (SKVR, JR) runosong corpora.
Translation index: 471,241 English → Finnic lemma mappings bridging MetaNet’s English keywords to the Finnic vocabulary.
Thesaurus: 25 semantic domains with 114K categorized lemmas providing domain-level structure for cross-domain analysis.

Three matching tiers

Tier 1 — Substitution Evidence (primary, green badge): MetaNet frame keywords are translated to Finnic lemmas, then checked against substitution pairs from parallel verses. When a source-domain word and target-domain word appear as a substitution pair, this is direct corpus evidence of the metaphor. Common function words (olema, minema, ära, etc.) are excluded to avoid false positives. Thresholds: Jaccard ≥ 0.1, co-occurrence ≥ 3, semantic type only.
Tier 2 — Domain Crossing (secondary, yellow badge): Bottom-up discovery — counts substitution pairs that cross between thesaurus domains, then maps these crossings to MetaNet metaphors. Shown in the Domain tab as a ranked table.
Tier 3 — Semantic Similarity (fallback, gray badge): GloVe-300 word embedding cosine similarity between metaphor names and thesaurus domain centroids. Lower confidence — proves word relatedness but not that the metaphor is active in the corpus.

How to use

By Family (default tab): Browse metaphor families sorted by match count. Click a family header to expand and see its members. Each metaphor card shows the source → target frame mapping, a tier badge, score, and language indicator. Click any matched metaphor to see its full evidence panel with lemmas, substitution pairs, and verse examples.
By Domain: View the cross-domain substitution table showing the 50 strongest domain crossings out of 143 found. Click any row to expand all pairs with verse examples. These represent the corpus’s own metaphorical structure — domains that substitute for each other in parallel verses.
Search: Type ≥2 characters to search metaphor names, frame names, matched lemmas, and English keywords. Works for both English terms (“fire”) and Finnic lemmas (“tamm”).

Filters

All: Show all families and metaphors (including unmatched, shown grayed out).
Matched only: Only metaphors with corpus evidence (any tier).
Tier 1 / Tier 2 / Tier 3: Filter by specific matching tier.
Estonian / Finnish / Both ET+FI: Filter by the language of substitution evidence. “Both ET+FI” shows metaphors with evidence in both traditions — particularly interesting for cross-cultural comparison.

Scores & confidence

Tier 1 score = evidence_score × 10 + keyword_score, where evidence_score = ∑(Jaccard × log(1 + co-occurrence)) over all matched substitution pairs. Higher scores mean more and stronger substitution evidence. Scores above 50 (green) indicate robust corpus evidence; 20–50 (yellow) moderate; below 20 (gray) weak.
Tier 3 score = cosine similarity × 10. These are semantic word-level matches only, not corpus evidence.
Tier badges: T1 strong evidence, T2 domain-level, T3 semantic fallback.

Reading the evidence panel

Click any matched metaphor card to expand its evidence panel:
Source/Target Lemmas: Finnic words linked to MetaNet’s source and target frames, clickable to the Lexicon. These are the vocabulary that the metaphor operates through in runosong.
Substitution pairs: Word pairs from parallel verses where one word belongs to the source domain and the other to the target domain. Each pair shows Jaccard similarity (J), co-occurrence count (n), language, and an example verse pair from the corpus.
Verse examples: Actual parallel verses from ERAB (Estonian) or SKVR/JR (Finnish) showing the substitution in context, with the source reference in parentheses.

URL deep linking

?q=fire — search for “fire”
?metaphor=ANGER_IS_FIRE — jump to a specific metaphor
?tab=domains — open the domain crossing tab
?tab=search — open the search tab

Limitations

MetaNet is English-centric — matching depends on English translations of Finnic vocabulary, so culture-specific metaphors without English equivalents may be missed. Substitution evidence is indirect: a pair proves structural equivalence in parallel verses, which is necessary but not sufficient for metaphor (some pairs may reflect metonymy or other relations). Tier 3 (GloVe) matches indicate word-level similarity only, not metaphorical activity in the corpus. Only ~2,000 lemmas have thesaurus domain labels, limiting Tier 2 coverage.

Attribution

MetaNet: Dodge et al. (2015), Stickles et al. (2016), Berkeley ICSI / UBC. No explicit license; academic fair use for research. Conceptual Metaphor Theory: Lakoff & Johnson (1980). Substitution data from RunoVerse parallel verse analysis. GloVe embeddings: Pennington, Socher & Manning (2014). Runosong parallelism scholarship: Frog (2016), Sarv (2017), Tarkka (2005).