diff --git a/frontend/src/views/social/VocabLessonView.vue b/frontend/src/views/social/VocabLessonView.vue
index 21b8466..72651b7 100644
--- a/frontend/src/views/social/VocabLessonView.vue
+++ b/frontend/src/views/social/VocabLessonView.vue
@@ -1318,39 +1318,47 @@ export default {
       // damit weder leere Gloss-Zeilen noch übergroße Listen entstehen.
       const out = [];
       const seen = new Set();
-      let nativeHints = new Set(
-        (this.normalizedCorePatterns || [])
-          .map((x) => this.normalizeLessonVocabTerm(x?.gloss))
-          .filter(Boolean)
-      );
-      let targetHints = new Set(
-        (this.normalizedCorePatterns || [])
-          .map((x) => this.normalizeLessonVocabTerm(x?.target))
-          .filter(Boolean)
-      );
-      if (nativeHints.size === 0 || targetHints.size === 0) {
-        nativeHints = new Set(
-          (this.importantVocab || [])
-            .map((x) => this.normalizeLessonVocabTerm(x?.learning))
-            .filter(Boolean)
-        );
-        targetHints = new Set(
-          (this.importantVocab || [])
-            .map((x) => this.normalizeLessonVocabTerm(x?.reference))
-            .filter(Boolean)
-        );
-      }
+      const targetTokenWeight = new Map();
+      const nativeTokenWeight = new Map();
+      const addTokens = (text, map, weight = 1) => {
+        const tokens = String(text || '')
+          .toLowerCase()
+          .normalize('NFKC')
+          .replace(/[\p{P}\p{S}]+/gu, ' ')
+          .split(/\s+/)
+          .map((t) => t.trim())
+          .filter((t) => t.length >= 2);
+        tokens.forEach((token) => {
+          map.set(token, (map.get(token) || 0) + weight);
+        });
+      };
+      // Core patterns gelten als qualitativ beste Quelle -> höheres Gewicht
+      (this.normalizedCorePatterns || []).forEach((p) => {
+        addTokens(p?.target, targetTokenWeight, 3);
+        addTokens(p?.gloss, nativeTokenWeight, 3);
+      });
+      // Übungs-Extraktion als Zusatzsignal
+      (this.importantVocab || []).forEach((v) => {
+        addTokens(v?.reference, targetTokenWeight, 1);
+        addTokens(v?.learning, nativeTokenWeight, 1);
+      });
+      const sideScore = (text, map) => {
+        const tokens = String(text || '')
+          .toLowerCase()
+          .normalize('NFKC')
+          .replace(/[\p{P}\p{S}]+/gu, ' ')
+          .split(/\s+/)
+          .map((t) => t.trim())
+          .filter((t) => t.length >= 2);
+        return tokens.reduce((sum, token) => sum + (map.get(token) || 0), 0);
+      };
       const orientPair = (target, gloss) => {
         const t = String(target || '').trim();
         const g = String(gloss || '').trim();
         if (!t || !g) return { target: t, gloss: g };
-        const nt = this.normalizeLessonVocabTerm(t);
-        const ng = this.normalizeLessonVocabTerm(g);
-        const tLooksNative = nativeHints.has(nt) && !targetHints.has(nt);
-        const gLooksTarget = targetHints.has(ng) && !nativeHints.has(ng);
-        const tLooksTarget = targetHints.has(nt) && !nativeHints.has(nt);
-        const gLooksNative = nativeHints.has(ng) && !targetHints.has(ng);
-        if (tLooksNative && gLooksTarget && !(tLooksTarget && gLooksNative)) {
+        const directScore = sideScore(t, targetTokenWeight) + sideScore(g, nativeTokenWeight);
+        const swappedScore = sideScore(g, targetTokenWeight) + sideScore(t, nativeTokenWeight);
+        if (swappedScore > directScore) {
           return { target: g, gloss: t };
         }
         return { target: t, gloss: g };